Assessment of Survey Data for the Analysis of Marriage and Divorce

THE URBAN INSTITUTE 2100 M STREET, N.W. / WASHINGTON D.C. 20037

ASSESSMENT OF SURVEY DATA FOR THE

NALYSIS OF MARRIAGE AND DIVORCE AT

THE

NATIONAL, STATE, AND LOCAL LEVELS

REPARED FOR:

U.S. D

EPARTMENT OF HEALTH AND HUMAN SERVICES,

FFICE OF THE ASSISTANT SECRETARY FOR PLANNING AND EVALUATION

DMINISTRATION FOR CHILDREN AND FAMILIES

REPARED BY:

AROLINE RATCLIFFE

REGORY ACS

IM DORE

AVID MOSKOWITZ

EPTEMBER 2008

The views expressed by the authors do not necessarily reflect the official policies of the

Department of Health and Human Services nor does mention of trade names, commercial

practices, or organizations imply endorsement by the U.S. government

Acknowledgments

The authors wish to thank staff at the National Center for Health Statistics and the U.S. Census

Bureau for providing valuable information about key data sets reviewed in this study. Members

of the expert panel, as well as staff from the U.S. Department of Health and Human Services

and The Lewin Group provided helpful comments and suggestions.

Table of Contents

EXECUTIVE SUMMARY .........................................................................................................................i

I. INTRODUCTION ..........................................................................................................................1

II. CRITERIA FOR EVALUATION ..................................................................................................2

Relevancy: Survey Data Can Be Used to Calculate Marriage and Divorce Rates ................ 2

Reliability: Survey Design Likely to Provide Estimates of Marriage and Divorce Rates

Representativeness: Survey Data Capture Broad U.S. Population and Provide State

Contains Correlates of Interest: Survey Includes Correlates and Outcomes of Interest to

that Match an External Source...............................................................................................3

and/or Local Level Estimates................................................................................................5

Ongoing: Survey Planned to Continue into Foreseeable Future ............................................6

Research and Policy Community.......................................................................................... 7

Summary.........................................................................................................................................7

III. KEY DATA SETS IDENTIFIED AND THE CRITERIA THEY MEET .................................7

Relevancy: Survey Data Can Be Used to Calculate Marriage and Divorce Rates ................ 8

Reliability: Survey Design Likely to Provide Estimates of Marriage and Divorce Rates

Representativeness: Survey Data Capture Broad U.S. Population and Provide State

Contains Correlates of Interest: Survey Includes Correlates and Outcomes of Interest to

that Match an External Source...............................................................................................8

and/or Local Level Estimates................................................................................................9

Ongoing: Survey Planned to Continue into Foreseeable Future ............................................9

Research and Policy Community........................................................................................ 10

Comparison of ACS, SIPP, and NSFG ......................................................................................10

IV. DETAILS OF THE THREE KEY DATA SETS......................................................................... 11

American Community Survey (ACS) .......................................................................................12

National Survey of Family Growth (NSFG) ............................................................................14

Survey of Income and Program Participation (SIPP) ............................................................. 19

Summary.......................................................................................................................................23

V. REFERENCES ................................................................................................................................25

VI. APPENDIX A: OTHER DATA SETS EXAMINED.................................................................27

VII. APPENDIX B: MARRIAGE AND DIVORCE QUESTIONS................................................ 35

EXECUTIVE SUMMARY

Whether they are assessing options for reforming Social Security, contemplating changes to the

tax code, or trying to improve the well-being of children, policy makers require an accurate

assessment of trends in marriage, divorce, and other living arrangements. Historically, vital

statistics data collected at the state and local levels and consolidated by the National Center for

Health Statistics (NCHS) provided the most complete information on marriage and divorce

rates. These data were based on administrative records of actual marriages and divorces that

occurred in the reporting jurisdictions. In 1996, however, NCHS discontinued funding to states

for the collection of detailed marriage and divorce data; NCHS continues to collect counts.

The Administration for Children and Families and the Office of the Assistant Secretary

for Planning and Evaluation, within the U.S. Department of Health and Human Services

(DHHS) contracted with the Lewin Group and the Urban Institute to explore options for

collecting marriage and divorce information. This report examines the feasibility and potential

benefits of using existing survey data sets to provide reliable, timely information on marriage

and divorce. It assesses the ability of a variety of data sets to produce marriage and divorce

statistics at the national, state, and local levels. The main criterion is whether the existing survey

data sets provide or can be modified to provide information on marriage and divorce rates, as

was collected under the vital statistics system.

To identify survey data sets that have the greatest potential for collecting marriage and

divorce statistics, the research team established five evaluation criteria. These criteria are used

to assess the surveys’ overall relevance and potential for providing marriage and divorce rates

over time. The criteria are: (1) relevancy—survey data can be used to calculate marriage and

divorce rates, (2) reliability—survey design is likely to provide estimates of marriage and

divorce rates that match an external source, (3) representativeness—survey captures broad U.S.

population and survey provides state and/or local level estimates, (4) ongoing—survey is

planned to continue into foreseeable future, and (5) contains correlates of interest—survey

includes correlates and outcomes of interest to research and policy communities. Based on these

criteria, three data sets are identified as having the greatest potential for measuring marriage

and divorce statistics. These data sets are:

1. The American Community Survey (ACS)

2. The National Survey of Family Growth (NSFG)

3. The Survey of Income and Program Participation (SIPP)

A summary of these three primary data sets is provided in Exhibit 3 of the report and the

questions that these three data sets ask about marriage and divorce are presented in Appendix

B. In addition, Appendix A summarizes information about the 17 other data sets examined but

determined not to be strong data sets for measuring marriage and divorce. Selected information

about the three key data sets is presented below.

American Community Survey (ACS)

• The ACS is a cross-sectional survey that is replacing the Decennial Census long form

and is planned to continue annually. The most recent ACS data available is for 2006.

• Prior to the 2008 questionnaire, the ACS had a single question on current marital status,

so could not be used to calculate marriage and divorce rates. Beginning in 2008, the ACS

questionnaire includes retrospective questions on marriage and divorce that can be used

to calculate marriage and divorce rates.

• State-level marriage and divorce rates can be computed for all 50 states and the District

of Columbia on an annual basis. Estimates for most counties are available using three

years of pooled data and Census tract-level estimates are available with five years of

pooled data.

• The survey samples approximately 3 million addresses per year and includes persons

living in military barracks and institutions.

• The ACS provides information on the demographic characteristics, program

participation, and economic status of households, although the correlates are more

limited set than what is available in the SIPP and NSFG.

National Survey of Family Growth (NSFG)

• The NSFG is a cross-sectional survey that provides retrospective information on

marriage, divorce, and cohabitation. These retrospective histories allow marriage and

divorce rates to be calculated.

• The NSFG includes the civilian noninstitutionalized population between the ages of 15

and 44, and can provide estimates of marriage and divorce rates for this population at

the national and regional levels (four Census regions). The NSFG can be used to

calculate national marriage and divorce rates but cannot be used to calculate marriage

and divorce rates at the state or local level.

• NSFG data are currently available from six cycles (1973, 1976, 1982, 1988, 1995, and 2002)

and a seventh cycle is underway (2006-2010). The first data from Cycle 7 is expected to

be released in late 2009.

• The sample size for the 2002 NSFG sample was 12,600. Cycle 7, which will be

administered continuously from 2006 through 2010, is expected to have a total sample

size of 20,000, with an average of 5,000 interviews per year.

• The NSFG provides information on many correlates of marriage, divorce, and

cohabitation. This includes questions on attitudes toward marriage, parenthood, and

gender roles, as well as family planning practices, pregnancy, and birth histories.

Survey of Income and Program Participation (SIPP)

• The SIPP is a longitudinal data set that also collects retrospective information on

marriage and divorce, and thus, marriage and divorce rates can be calculated.

• Each SIPP panel is a nationally representative sample of the U.S. civilian non-

institutionalized population. The SIPP can only provide reliable estimates of marriage

and divorce rates at the national level and for the four Census regions.

• SIPP panels began annually from 1984 to 1993 and in 1996, 2001, and 2004; panels are

typically two to four years in length. A 2008 panel is planned and a re-design of the SIPP

instrument is underway.

• The SIPP only provides reliable estimates of marriage and divorce rates once in each

panel, or every three to five years

• The 2004 SIPP panel is the largest panel to date and includes 46,500 households. The

1996 and 2001 SIPP panels include 36,800 and 35,100 households, respectively.

• The SIPP provides a wide variety of correlates beyond demographic characteristics,

including benefit recipiency, employment, and fertility histories, as well as child and

adult well-being.

iii

I. INTRODUCTION

Whether they are assessing options for reforming Social Security, contemplating changes to the

tax code, or trying to improve the well-being of children, policy makers require an accurate

assessment of trends in marriage, divorce, and other living arrangements. Historically, vital

statistics data collected at the state and local levels and consolidated by the National Center for

Health Statistics (NCHS) provided the most complete information on marriage and divorce

rates.

These data were based on administrative records of actual marriages and divorces that

occurred in the reporting jurisdictions. Prior to 1996, the federal government provided more

financial support to states to collect marriage and divorce data. In 1996, NCHS discontinued

funding for the collection of detailed marriage and divorce data for a number of reasons.

First,

there were resource constraints. By ending funding to states for the collection of detailed

marriage and divorce data, the agency was able to redirect $1.25 million per year to maintaining

birth and death data systems. Second, there were coverage and quality concerns. NCHS was not

consistently receiving data from all states. Similarly, the quality of the data never rose to the

level expected by NCHS or researchers. Data reporting was often incomplete or of uncertain

reliability. States were facing staffing shortages and internal funding issues, resulting in many

relegating the reporting and collection of marriage and divorce data to a lower priority than

birth and death data. Finally, the collection of marriage and divorce information was an uneasy

fit for an agency that focuses on health statistics. Few NCHS staff worked with marriage and

divorce data; thus, when the budget needed to be cut, there were few advocates for the

continuation of its collection. In addition, staff argued that the data could be captured in

surveys, such as the National Survey of Family Growth, and in Census data. As a result, the

quantity and quality of national, state, and local information on marriage and divorce

deteriorated while policy makers’ need for this information increased.

The Administration for Children and Families and the Office of the Assistant Secretary

for Planning and Evaluation, within the U.S. Department of Health and Human Services

(DHHS) contracted with the Lewin Group and the Urban Institute to explore options for

collecting marriage and divorce information. This report examines the feasibility and potential

benefits of using existing survey data sets to provide reliable, timely information on marriage

and divorce.

It assesses the ability of a variety of data sets to produce marriage and divorce

statistics at the national, state, and local levels. The main criterion is whether the existing survey

data sets provide or can be modified to provide information on marriage and divorce rates, as

was possible using data collected under the vital statistics system.

Beyond their potential to provide information on marriage and divorce rates, survey

data sets can also provide information on current marital status, alternative living arrangements

such as cohabitation, and the correlates of marriage, divorce, and other living arrangements.

This report first describes criteria for evaluating survey data sets on their ability to

provide information on marriage and divorce. It then highlights three data sets that were

NCHS is part of the Centers for Disease Control and Prevention, an agency within the U.S. Department of Health and Human

Services (DHHS).

NCHS still provides states with a small amount of financial support to collect and send to NCHS basic counts of marriages and

divorces.

A separate project report explores the vital statistics system. Lewin Group. Collection of Marriage and Divorce Statistics by

States. Final report to U.S. Department of Health and Human Services. April 2008.

identified as being most likely to meet these criteria and capture marriage and divorce

information, although some require modifications. These three data sets are:

1. The American Community Survey (ACS)

2. The National Survey of Family Growth (NSFG)

3. The Survey of Income and Program Participation (SIPP)

Many other data sets have merit but do not adequately meet the criteria established for

this assessment; information on these data sets, along with the criteria they failed to meet,

appears in Appendix A. Within the presentation of the three preferred data sets, the report

discusses potential modifications that could improve their usefulness.

II.

CRITERIA FOR EVALUATION

To identify survey data sets that have the greatest potential for collecting marriage and divorce

statistics, we establish five evaluation criteria. These criteria are used to assess the surveys’

overall relevance and potential for measuring marriage and divorce. Failure to meet all five

criteria does not imply the data are not useful for measuring marriage and divorce. Rather, we

use information gathered from these assessments to explore ways to modify the data sets to best

meet these criteria. The criteria are:

1. Relevancy: Survey data can be used to calculate marriage and divorce rates

2. Reliability: Survey design is likely to provide estimates of marriage and divorce rates

that match an external source

3. Representativeness:

• Survey captures broad U.S. population

• Survey provides state and/or local level estimates

4. Ongoing: Survey is planned to continue into foreseeable future

5. Contains correlates of interest: Survey includes correlates and outcomes of interest to

research and policy communities

Below, each of these criteria and the justification for considering them is discussed.

Relevancy: Survey Data Can Be Used to Calculate Marriage and Divorce Rates

The U.S. vital statistics system provided marriage and divorce rates, so this criterion evaluates

the extent to which survey data can provide this information. Calculating marriage and divorce

rates (the number of marriages and divorces per 1,000 people in a particular year) requires

information on marital status over time. Specifically, marriage rates are calculated by

identifying individuals who transition from being not married (last year) to married (this year),

and divorce rates are calculated by identifying individuals who were married last year but are

divorced this year. Alternatively, these rates can be calculated if the survey asks individuals

about their marital histories (i.e., if and when they married and divorced). Surveys that only

provide information on marital status at a single point in time cannot be used to calculate

marriage and divorce rates; however, these data do provide important information about the

marital status of the U.S. population (e.g., the percent currently married or currently divorced).

Whether and how a survey can be used to calculate marriage and divorce rates depends

on its design. Survey data can be grouped into four design categories: (1) cross-sectional (i.e.,

single point in time) surveys that do not include retrospective relationship information; (2)

cross-sectional surveys that include retrospective relationship information; (3) longitudinal (or

panel) surveys that track the same individuals over time but provide no retrospective

information; and (4) longitudinal surveys that also contain retrospective relationship data.

These four types of survey data vary in their ability to provide estimates of individuals' marital

status at a point in time, transitions in and out of marital (and cohabiting) relationships, and the

marital and relationship histories of individuals.

In general, cross-sectional surveys that do not ask about relationship histories can

capture marital and cohabitation status at a point in time but provide no information on

changes in marital status. Thus, marriage and divorce rates cannot be calculated. This is a

significant drawback of these data. In contrast, cross-sectional surveys that specifically ask

about past marriages and divorces provide information on both current marital status and

changes in status, allowing marriage and divorce rates to be calculated.

Longitudinal data sets that ask about marital status multiple times during the panel can

capture current relationship status as well as changes that occur during the time survey

respondents are tracked. These longitudinal data, however, provide only a limited window of

time in which transitions are captured (i.e., the duration of the panel). Finally, longitudinal data

with retrospective information capture current status as well as past and present changes in

relationship status.

Thus, longitudinal data sets and data sets that include retrospective marriage and

divorce data provide information required to calculate marriage and divorce rates and best

satisfy the “relevancy” criterion.

Reliability: Survey Design Likely to Provide Estimates of Marriage and Divorce

Rates that Match an External Source

This section examines the reliability of estimates of marriage and divorce rates based on

longitudinal and retrospective data. Exhibit 1 summarizes the advantages and disadvantages of

retrospective versus longitudinal data for examining marital transitions and histories.

Retrospective information on marital events generally coincides with information

collected in longitudinal data. For example, Lillard and Waite compare the 1985 Panel Study of

Income Dynamics (PSID) retrospective data with marital events over the previous 17 years of

panel data and find significant agreement between the two.

Retrospective histories require

individuals to recall past events while longitudinal surveys capture events shortly after they

occur. Although some have argued that retrospective data are prone to recall error and may

systematically under-record brief status changes and embarrassing events,

research shows that

As noted above, some data sets include both retrospective and longitudinal data.

Lillard, L.A., and L. J. Waite. 1989. “Panel versus Retrospective Data on Marital Histories: Lessons from the PSID.” In Individuals

and Families in Transition: Understanding Change through Longitudinal Data (243-53). Washington, DC: U.S. Bureau of the Census.

Bumpass, L., K. Raley. 2007. “Measuring Separation and Divorce.” In Handbook of Measurement Issues in Family Research, edited

by Sandra L. Hofferth and Lynne M. Casper (125–44). Mahwah, NJ: Lawrence Erlbaum Associates.

these data do a fairly good job of capturing marital history. For example, Krieder and Fields

compare vital statistics data to retrospective histories collected in a topical module of the 1996

SIPP and find that the SIPP captures 97-98 percent of the total recent marriages.

7,8

In addition,

research suggests the histories were still relatively accurate 10 to 15 years in the past; however,

retrospective divorce data are not as accurate as marriage data.

Further, retrospective

information on the formation and dissolution of cohabitating unions is expected to be less

reliable than changes in marital status.

Exhibit 1. Advantages and Disadvantages of Retrospective versus Longitudinal Data

Type of Data

Retrospective Data

•

Advantages

Capture more marital events over longer

periods of time than longitudinal data

Covers a more representative sample than

longitudinal data

•

Disadvantages

May be prone to recall error

Underreports brief status changes or

sensitive information

Data not as accurate for measuring

divorce as marriage

Less reliable data on formation and

dissolution of cohabitating unions

than changes in marital status

Longitudinal Data

•

Captures events as they occur

Measures correlates of marital events at

time of occurrence

Useful for examining causes and

consequences of changes in marital status

May be more reliable than retrospective

data for capturing changes in cohabitation

•

Attrition bias

Marital status may change multiple

times between waves of a panel

Does not capture complete marital

histories

There are several drawbacks to using longitudinal (i.e., panel) data to study marital

events. First, the participants who drop out of longitudinal studies may be more unstable than

those who remain, leading to attrition bias.

It is not uncommon for individuals to exit panel

surveys at a time of separation or divorce, leading to under-representation of marital status

changes. Evidence of this comes from the 1986 panel of the SIPP where the divorce/separation

rate of couples when neither spouse exited the survey was around 2 percent compared with 60

Lillard, L.A., and L. J. Waite. 1989. “Panel versus Retrospective Data on Marital Histories: Lessons from the PSID.” In Individuals

and Families in Transition: Understanding Change through Longitudinal Data (243-53). Washington, DC: U.S. Bureau of the Census.

Kreider, R. M., and J. M. Fields. 2002. “Number, Timing, and Duration of Marriages and Divorces: 1996.” Current Population

Reports, Series P-70, No. 80. Washington, DC: U.S. Bureau of the Census.

The Survey of Income and Program Participation is discussed in detail in Section IV.

Kreider, R. M., and J. M. Fields. 2002. “Number, Timing, and Duration of Marriages and Divorces: 1996.” Current Population

Reports, Series P-70, No. 80. Washington, DC: U.S. Bureau of the Census.

Manning W., and P. Smock. 2003. “Measuring and Modeling Cohabitation: New Perspective from Qualitative Data.” Center for

Family and Demographic Research Working Paper #2003-10. Their findings suggest that because cohabitation is sometimes not

a deliberate decision, there may not be a clear beginning and end date of cohabitation, and couples may have a difficult time

remembering when they started cohabiting.

Lillard, L.A., and L. J. Waite. 1989. “Panel versus Retrospective Data on Marital Histories: Lessons from the PSID.” In Individuals

and Families in Transition: Understanding Change through Longitudinal Data (243-53). Washington, DC: U.S. Bureau of the Census.

percent for longitudinal panel members where one spouse left the panel.

Similarly,

longitudinal data sets may become less representative of the overall population after the initial

year as the population changes due to factors such as immigration. Another drawback of panel

surveys that only ask current marital status is that marital status may change multiple times

between waves of a panel study. Finally, longitudinal data only span the time in the panel and

cannot cover individuals’ complete marital histories. It is important to note, however, that panel

data may be preferred to retrospective data in the case of cohabitation, as individuals' recall

related to when cohabiting unions began and ended might be less accurate than their

recollection about when a marriage or divorce occurred.

In general, retrospective histories can capture more marital events over longer time

periods and for more representative samples of the population than longitudinal data.

Longitudinal data, however, measure the correlates of marital events at the time they occur and

are more useful for examining the causes and consequences of changes in marital status than

retrospective histories.

Representativeness: Survey Data Capture Broad U.S. Population and Provide State

and/or Local Level Estimates

Ideally, survey data would provide marriage and divorce rates for the entire U.S. population, as

well as for states and localities. However, even broad-based national surveys may exclude some

populations from their sample universes (e.g., persons in institutions), and others may target

specific population subgroups (e.g., adolescents). The evaluation of this criterion considers the

degree to which the sample population is restricted to a subset of the total U.S. population. For

example, a survey that includes individuals' ages 15 to 45 is more useful for providing marriage

and divorce data and has a greater potential for modification (i.e., it would likely be easier and

more straightforward to modify) than a survey that focuses on adolescents only.

In addition to the particular population selected for the survey, a second issue arises

when considering longitudinal data sets: the extent to which these data continue to represent

the U.S. population in the years after the sample population is initially drawn. For example, a

sample population drawn in the 1960s will not be representative of the current U.S. population

because of compositional shifts, such as the change in the immigrant population (both in terms

of number and country of origin). For this reason, long panels that do not include refresher

samples to account for changes in the composition of the U.S. population will not meet this

criterion.

Finally, the goal is to have survey data that provide marriage and divorce statistics at the

national, state, and local levels. To calculate state-level marriage and divorce rates, for example,

it is not enough for a data set to simply identify individuals' state of residence. Rather, the

sample size and sample design must be such that the data set has a sample that is

Hill, D. H. 1997. “Adjusting for Attrition in Event-History Analysis.” Sociological Methodology 27: 393–416. Further information

on attrition-related biases in the PSID and SIPP can also be found in Becketti et al. (1988), Fitzgerald, Gottschalk, and Moffitt

(1998), Lillard and Panis (1998), and Zabel (1998).

There are other issues that could be considered here, such as the survey response rates and the permissibility and use of proxy

responses. These issues are addressed in the discussion of selected data sets in Section IV below.

Surveys that are not broadly representative can, nonetheless, be useful for examining outcomes for specific population sub-

groups (e.g., adolescents, a single birth cohort, or new parents).

"representative" of the state’s population. The same is true for other geographic areas. Further,

the goal is to have data sets that provide information on a large portion of the U.S. population.

A data set that is specific to one state or selected cities, for example, is not particularly useful for

measuring marriage and divorce statistics to the extent that vital statistics once provided these

data.

Ongoing: Survey Planned to Continue into Foreseeable Future

The goal of this study is to identify survey data sets that can be used to calculate marriage and

divorce statistics in the coming years. Focusing on ongoing surveys and identifying ways in

which these surveys can be modified to provide the information of interest is considerably less

expensive than proposing new survey efforts. However, it is also useful to review discontinued

surveys because they help identify ways to modify ongoing survey efforts to reach the data

goals. Ideally, the data sets would provide information on an annual or biennial basis, and there

would not be a long lag between data collection and release.

Surveys implemented on an ongoing basis provide information from which to evaluate

changes in marriage rates over time. However, marriage and divorce are relatively rare events,

so detecting small changes over time requires relatively large samples. The annual marriage rate

for males and females ages 15 and older, for example, is roughly 25 per 1,000 persons (or 2.5

percent).

With an annual sample size of 10,000, one has an 80 percent chance of detecting a

change in the percent of people married of at least 0.56 percentage points.

In other words, one

can expect that samples will lead to a statistically significant difference when the true change is

from 2.5 percent to 3.1 percent, but less confident that they will reveal a significant difference

when the true change is smaller (e.g., from 2.5 percent to 2.8 percent). With an annual sample

size of 100,000, one can be confident of detecting changes of at least 0.18 percentage points.

Thus, a large sample is important if a goal is to detect changes over time.

These calculations do not imply that marriage rates calculated with modest sample sizes

are not useful, only that it is difficult to detect changes over time with modest sample sizes. The

precision of the estimated marriage rate within a year is much greater. If the marriage rate is

calculated with a survey sample size of 10,000 and the estimated percentage of people who get

married is 2.5 percent, then one can be 90 percent confident that the true percentage falls within

0.26 percentage points of the estimated value—or between 2.24 percent and 2.76 percent.

Clarke, S.C. Advance Report of Final Marriage Statistics, 1989 and 1990. Monthly Vital Statistics Report 1995; 43(12): 7. The

numbers reported are for 1990—the marriage rate for males and females ages 15 and older were 26.0 and 24.1 per 1,000 persons,

respectively.

These calculations are based on two criterions that are commonly used by statisticians to assess a proposed sample size. First,

the significance level is set at 10 percent. That is, if no true change exists, one should find statistically significant differences less

than 10 percent of the time. A two-tailed test is used. Second, the “power” is set to 80 percent. That is, if a true difference exists,

we require at least an 80 percent chance of finding a statistically significant difference in the sample. These two criteria are then

used to calculate the smallest change for which our proposed sample has at least an 80 percent chance of finding “statistically

significant” difference. Note that these calculations are based on the Poisson distribution and assume that the sample is

randomly drawn.

Similar issues arise if one was evaluating differences in marriage rates across states.

Contains Correlates of Interest: Survey Includes Correlates and Outcomes of Interest

to Research and Policy Community

In addition to having data to calculate marriage and divorce rates at the national, state, and

local level, survey data would ideally include information necessary to calculate these rates for

different subpopulations (such as by age, gender, race/ethnicity, immigration status, and urban

versus rural). As a result, this criterion considers the extent to which survey data sets include

these characteristics. Beyond this, many in the research and policy community are interested in

understanding and researching the relationship between marriage and divorce and outcomes

such as participation in public assistance programs, families' economic and material well-being,

and children's well-being. With these data, for example, one could study the impact of divorce

on children's well-being. For this reason, this review considers the extent to which data sets

include characteristics of families and children that may affect or be affected by marital status

and changes in marital status. Finally, data sets that measure marital and relationship quality

are examined. However, since having detailed correlates is not important for calculating

marriage and divorce rates, this criterion is considered a secondary criterion.

Summary

The five evaluation criteria—relevancy, reliability, representativeness, ongoing, and containing

correlates of interest—serve as a framework for assessing how effectively survey data sets can

be used to compute information comparable to vital statistics and to analyze the correlates and

consequences of marriage and divorce. The next section discusses data sets identified as having

the greatest potential for providing this information.

III.

KEY DATA SETS IDENTIFIED AND THE CRITERIA THEY MEET

Three data sets have been identified as having the greatest potential for measuring marriage

and divorce statistics. As noted above, these data sets are the American Community Survey

(ACS), the National Survey of Family Growth (NSFG), and the Survey of Income and Program

Participation (SIPP). Until recently, none of these data sets met all of the criteria discussed

above, nor did they provide data comparable to the pre-1996 vital statistics system. However,

with the implementation of changes to the ACS in 2008, the ACS now meets all five criteria.

While the NSFG and SIPP do not meet all five criteria, they are important data sets for

understanding marriage and divorce. Below, these three data sets are discussed in the context of

the five criteria presented above.

Exhibit 2 identifies the criteria that the three data sets meet.

More specific information about each of these data sets is presented in Section IV below.

All of the surveys examined collect information on basic demographic characteristics (e.g., age, gender, race/ethnicity), which

are necessary to calculate marriage and divorce rates for subpopulations.

Appendix A identifies and presents information on 17 other data sets reviewed for this study. These data sets were determined

to be relatively weak candidates for providing marriage and divorce statistics comparable to the former vital statistics system.

Exhibit 2: Evaluation Criteria Attained by Key Data Sets

Survey Data Set Relevancy Reliability

Representativeness

Ongoing Correlates

Broad U.S. State/Local

Estimates

Population

American Community Survey

(ACS)

x x x x x x

National Survey of Family

Growth (NSFG) x x x x

Survey of Income and

Program Participation (SIPP) x x x x x

* An x indicates that the data set fulfills the criterion.

The 2004 SIPP panel provides a representative sample of the civilian non-institutionalized population for 30 states, but Census

staff caution against using the SIPP to calculate state-level marriage and divorce rates because of sample size concerns. Earlier

SIPP panels do not provide a representative sample for any state.

Relevancy: Survey Data Can Be Used to Calculate Marriage and Divorce Rates

All three data sets provide the information needed to calculate marriage and divorce rates. The

NSFG is a cross-sectional data set that includes retrospective information on marriage and

divorce, thereby allowing marriage and divorce rates to be calculated. The SIPP is a

longitudinal data set that also includes retrospective information on marriage and divorce, and

thus, marriage and divorce rates can be calculated. Prior to 2008, the ACS, which is a cross-

sectional survey, asked only a single question on current marital status, with no information on

prior status from which to calculate marriage or divorce rates. Beginning in 2008, however, the

ACS questionnaire includes retrospective questions on marriage and divorce that can be used to

calculate marriage and divorce rates.

In addition to providing information to calculate marriage and divorce rates, the NSFG

and SIPP, unlike the ACS, collect marital histories. These data provide information on marriage

duration and allow for analyses of how individuals’ transition into and out of marriage over

time. The NSFG, for example, has been used to examine the probability that a first marriage

dissolves, the probability of remarriage following a divorce, and the probability that a second

marriage dissolves.

Reliability: Survey Design Likely to Provide Estimates of Marriage and Divorce

Rates that Match an External Source

All three data sets collect retrospective information on marriage and divorce, and as discussed

above, research suggests that retrospective questions provide information that closely resembles

vital statistics, particularly for marriage rates. This information is only available from the ACS

beginning with the 2008 data, so currently available ACS data cannot be used to calculate

marriage and divorce rates. The SIPP's retrospective data is gathered early in the SIPP panel, so

that sample attrition is not a serious concern with these retrospective marriage and divorce

Bramlett, M.D., and W. D. Mosher. 2002. “Cohabitation, Marriage, Divorce, and Remarriage in the United States.” National

Center for Health Statistics. Vital Health Statistics 23(22). Available at

http://www.cdc.gov/nchs/data/series/sr_23/sr23_022.pdf (Accessed February 12, 2008).

data. The SIPP also collects marital status through its longitudinal panel; however, research

suggests that the higher rates of sample attrition among individuals who experience a marital

disruption lessens the reliability of marriage and divorce rates estimated from longitudinal

data.

Representativeness: Survey Data Capture Broad U.S. Population and Provide State

and/or Local Level Estimates

The ACS sampling design is superior to the NSFG and SIPP in meeting this criterion. Unlike

many surveys, the ACS sample includes persons living in military barracks and in institutions.

In addition, the ACS has a very large sample size—3 million households per year—and allows

for location-specific annual estimates of all areas with 65,000 or more people; this includes all 50

states and the District of Columbia. Further, estimates for most counties are available with three

years of pooled data and Census tract-level estimates are available with five years of pooled

data.

Each panel of the SIPP is a nationally representative sample of the U.S. civilian non-

institutionalized population, so the SIPP sample is slightly more restrictive than the ACS

sample. The 2004 SIPP panel is the largest panel to date and includes 46,500 households.

Historically, SIPP data could only provide representative estimates at the national level and for

the four Census regions.

The 2004 SIPP panel also provides a representative sample for 30

states, but Census staff caution against using the SIPP to calculate state-level marriage and

divorce rates. The state sample sizes are modest and marriage and divorce are relatively rare

events, so state-level marriage and divorce rates would be based on only a handful of marriages

and divorces.

The NSFG is also limited to the civilian non-institutionalized population, and its sample

is further restricted by age. The two most recent NSFG cycles (2002 and 2006-10) have the

broadest populations the survey has had to date and include men and women between the ages

of 15 and 44.

This restriction on age limits the extent to which the survey captures marriage

and divorce activity for the U.S. population. Further, the NSFG cannot be used to produce state-

and local-level estimates. It does, however, provide national-level estimates and estimates for

the four Census regions.

Ongoing: Survey Planned to Continue into Foreseeable Future

All three surveys—ACS, NSFG, and SIPP—are planned to continue into the future. The

frequency with which they will be administered and the speed with which their data will be

made available vary considerably across the three data sets. The ACS, which is replacing the

Decennial Census’s long form, provides annual data on 3 million households. The NSFG has

been administered intermittently in the past (1973, 1976, 1982, 1988, 1995, and 2002); however,

the current cycle of the NSFG (Cycle 7) has continuous interviewing from 2006 through 2010.

Although interviews will occur annually and each year of interviews will be a nationally

The four Census regions are Northeast, Midwest, South, and West. For more detailed information, see

http://www.census.gov/geo/www/us_regdiv.pdf (Accessed February 4, 2008).

In earlier cycles, the NSFG sample was also restricted by gender and marital status.

representative sample, NCHS staff (who administer the NSFG) recommend that a minimum of

two years of data be used to calculate estimates of marriage and divorce rates (because of

relatively small sample sizes—an average of 5,000 per year). NCHS expects that the NSFG will

be continuous, collecting data every year, indefinitely.

The most recent SIPP data available is from the 2001 and 2004 panels, which include

36,700 and 46,500 households, respectively.

A 2008 panel began in September 2008 and a re-

design of the SIPP instrument is also underway. Although the SIPP schedule over the last

decade is such that new panels have generally begun as old ones end, the key information on

retrospective marital histories is collected only once in each panel. Consequently, the SIPP only

provides reliable estimates of marriage and divorce rates once in each panel, or every three to

five years.

As discussed in Section II above, the size of the survey has implications for detecting

differences in marriage and divorce rates over time. The ACS has the largest sample size for

determining whether differences over time are statistically significant, followed by the SIPP and

then the NSFG.

Contains Correlates of Interest: Survey Includes Correlates and Outcomes of Interest

to Research and Policy Community

In terms of correlates and outcomes related to marriage and divorce, the SIPP has a great deal of

information related to program participation, economic and material well-being, and child well

being, among other individual and household characteristics. The NSFG also includes many

correlates where the focus is on the individual being surveyed, not the family or household.

Finally, the ACS provides information on program participation and economic status of the

household, although it contains a more limited set of correlates and outcomes than the SIPP and

NSFG. Section IV below details this information to a greater extent.

Comparison of ACS, SIPP, and NSFG

Currently available SIPP and NSFG data can be used to estimate marriage and divorce rates,

although neither data set can reliably provide this information for states or localities. While

currently available ACS data cannot be used for this purpose, it will be possible to calculate

marriage and divorce rates beginning with the release of the 2008 ACS. A significant advantage

of the ACS over the NSFG and SIPP is its ability to provide annual state-level estimates (for all

states plus D.C.), and local-level estimates with three to five years of pooled data.

The 2004 SIPP panel provides a representative sample for 30 states, but Census staff

caution against using the SIPP to calculate state-level marriage and divorce rates because the

state sample sizes are too small to accurately estimate the extent to which these relatively rare

events occur. The SIPP sample is small compared to the ACS—the 2004 SIPP sample is 46,500

households, whereas the ACS sample is 3 million households. Another drawback of the SIPP is

that reliable estimates of marriage and divorce rates are only available once during each panel

(currently every three to five years). The NSFG provides a viable alternative for measuring

All waves of the 2004 panel are not yet available

marriage and divorce rates at the national level on a continuous basis. However, as currently

designed, the NSFG only includes individuals between ages 15 and 44 and so does not capture

marriages and divorces for the full population.

IV.

DETAILS OF THE THREE KEY DATA SETS

The previous section discussed the extent to which the ACS, NSFG, and SIPP meet the five

criteria for measuring marriage and divorce. This section explores each data set in detail. For

each data set, a brief overview of the data set is presented, followed by a discussion of the

following areas: (1) current marriage, divorce, and cohabitation questions (Appendix B includes

a listing of marriage and divorce questions currently contained in the three surveys); (2) data

availability, access, and clarity; (3) marriage, divorce, and cohabitation correlates; (4) survey

response rate; (5) survey administration; and (6) process for modifying the survey. Exhibit 3

presents a summary of the three key data sets.

Exhibit 3: Summary of Key Data Sets

American Community

Survey (ACS)

Owner

• U.S. Census Bureau •

National Survey of Family

Growth (NSFG)

National Center for Health

Statistics

Survey of Income and

Program Participation (SIPP)

• U.S. Census Bureau

Survey Type

• Cross-sectional with

retrospective data

• Cross-sectional with

retrospective data

• Longitudinal with

retrospective data

Frequency

• Annual • 1973, 1976, 1982, 1988,

1995, 2002, 2006-2010

• 1984-1993, 1996, 2001,

2004

Population

• Total U.S. Population • Civilian non-institutionalized

population ages 15-44

• Civilian non-

institutionalized population

Sample Size

• 3,000,000 households • 12,600 individuals • 46,500 households

Mode

•

Mail-in • Face-to-face • Face-to-face and

telephone

Response Rate

• 97.5%

• 79% • 78.1%

Representative Areas

•

Nation

4 Census regions

All states

Local areas

•

Nation

4 Census regions

•

Nation

4 Census regions

30 states

Marriage/Cohabitation

Questions

•

Marital status

Cohabitation status of

reference person

Changes in marital

status in the past year

Number of marriages

Year of last marriage

•

Marital status

Cohabitation status

Marital history for all

marriages

Cohabitation history

•

Marital status

Cohabitation status

Marital history for up to 3

marriages

Unless otherwise noted, information on the ACS is based on the 2008 survey, the NSFG is based on Cycle 6 (2002), and the

SIPP is based on 2004 panel.

Households that do not return the mail-in questionnaire within one month are contacted for a telephone interview, and one-third of

those that do not respond to either the mail-in questionnaire or the telephone interview are contacted for a face-to-face interview.

The ACS response rate is from the 2006 survey.

This response rate is for Wave 2, when marital history data are collected. The Wave 1 response rate was 85.1 percent.

Estimates of areas with between 20,000 and 65,000 people (including most counties) can be computed by pooling three years of

data, and estimates of Census-tract areas with less than 20,000 people can be computed by pooling five years of data.

While the 2004 SIPP panel provides a representative sample for 30 states, Census staff caution against using the SIPP to

calculate state-level marriage and divorce rates because of sample size concerns.

American Community Survey (ACS)

The ACS is a cross-sectional survey that was designed to replace the Decennial Census long

form.

Like the Decennial Census, response to the ACS is required by law; it is the only survey

for which completion by the surveyed household is mandatory. The survey has been given in

test areas since 1996 and reached full implementation in 2005. In 2005, all households were

eligible to be surveyed, and starting in 2006 persons living in institutions and group quarters

(e.g., military barracks, prisons, college dormitories, homeless shelters, nursing homes) were

also eligible to be surveyed. The survey samples approximately 3 million households per year

(and 253,000 per month).

The ACS is the largest annual survey in the United States and is

designed to be representative down to the Census tract level. State-level statistics can be

computed for all 50 states and the District of Columbia on an annual basis. Annual statistics are

also available for all population areas with 65,000 people or more, which includes most

Metropolitan Statistical Areas (MSAs). Further, reliable county-level statistics can be computed

using three years of data and Census tract-level statistics can be computed using five years of

data.

For each household that completes an ACS questionnaire, one individual is identified as

the “household respondent.” The respondent is the person who owns or rents the house or

apartment and is the individual primarily responsible for completing the questionnaire.

The

ACS collects information on all household members.

Marriage, Divorce, and Cohabitation Questions. Prior to the 2008 questionnaire, the

ACS had a single question on marital status. This question identified an individual as married,

widowed, divorced, separated, or never married. As such, the ACS could not be used to

calculate marriage and divorce rates. This has been remedied, however, through the addition of

questions about changes in marital status to the 2008 questionnaire. These questions ask about

(1) change in marital status in the past 12 months, (2) number of times married, and (3) year last

married.

These additional questions are expected to remain in the survey for the foreseeable

future.

The ACS does not have a direct question about cohabitation status. Rather, cohabitation

status in the ACS is captured through a question about “relationship to respondent,” where one

of the allowable responses is “unmarried partner.” As a result, cohabitation status is only

available for survey respondents (and their partners). So, in cases where neither cohabiting

partner is the survey respondent, the cohabiting relationship is missed.

Data Availability, Access, and Clarity. Unlike many surveys, the main data product of

the ACS is a series of several hundred tables. These tables include information on marital status.

Tables based on one year of data are published for all geographic jurisdictions with populations

of 65,000 or more (e.g., states, towns, cities, American Indian reservations). With three years of

data, tables will be published for geographic jurisdictions with populations of 20,000 or more.

With five years of data, tables will be published down to the tract and block group level (similar

to what was published based on the Decennial Census long form data). These multi-year

Administration of the Decennial Census long form ended in 2000.

The ACS sample is spread across every county in the U.S. and includes all municipalities in Puerto Rico.

The ACS directions say that if no such person meets this criterion, the respondent can be any adult living in the household.

The exact ACS question can be found in Appendix B.

estimates will be released for the first time in 2008 and 2010. ACS data tabulations are available

through the Census Bureau's official data dissemination system, American FactFinder.

In addition to these tables, a public use data file is available for free download from the

Internet using DataFerrett.

For confidentiality purposes, the public use data file only identifies

areas with populations of 100,000 or more and is, at most, a one-third sample. Because of the

difference in samples, estimates calculated using this data file will not necessarily match the

published Census tables.

The tables and public use data file are typically released eight months after data

collection for each calendar year has ended. Because retrospective marriage and divorce data

were first collected in 2008, calculating marriage and divorce rates on geographic jurisdictions

with populations of 65,000 or more will first be possible with data released in summer/fall 2009.

Examining smaller jurisdictions require the data to be pooled across years, so marriage and

divorce rate data on geographic jurisdictions with populations of 20,000 or more will first

become available in 2011 and for Census tracts the data are expected to first become available in

2013. Data for all these units will then be released annually.

Like other Census data available on DataFerrett, the ACS data is very easy to download

and use and the documentation is clear and extensive.

Marriage, Divorce, and Cohabitation Correlates. The ACS provides information on

many demographic characteristics including age, race, ethnicity, immigration status, citizenship

status, and country of origin. The ACS also provides information on housing situations, such as

type of building, number of bedrooms, and whether there are complete plumbing and kitchen

facilities. In terms of public assistance programs, the ACS contains information on whether

anyone in the household received food stamps, as well as individuals’ receipt of welfare

benefits and Supplemental Security Income (during the past 12 months). The survey also

provides information on individuals’ employment, earnings, and total income (during the past

12 months). Finally, the ACS includes information on births in the last year.

Survey Response Rate. The ACS has a response rate of 97.5 percent (2006), the highest

of any Census Bureau demographic survey. Response rates have been found to vary by race

and ethnicity; households in minority-concentrated Census tracts have been found to have

significantly lower response rates. In the 2000 ACS, which had an overall response rate of 95

percent, the response rate in Census tracts where 75 percent of the population reported their

race as “black or African-American” was 92.4 percent, compared to 95.9 percent of tracts where

at least 75 percent were white.

This response rate of 92.4 percent in minority-concentrated

areas is still high compared to other survey response rates (including the NSFG and SIPP).

Survey Administration. The Census Bureau uses three modes of data collection —mail-

in questionnaires, telephone interviews, and in-person interviews—which contribute to high

The American Factfinder website is located at http://factfinder.census.gov.

DataFerrett is an application designed to create data extracts (for certain data sets) and to create tables without having to

download the entire raw data file. It is available for download through the Census website.

The difference between these rates is statistically significant. U.S. Bureau of the Census. 2002. “Meeting 21

Century

Demographic Data Needs—Implementing the American Community Survey: May 2002. Report 2: Demonstrating Survey

Quality.” Washington, DC: U.S. Bureau of the Census, p. 17.

response rates.

As a first step in the survey’s administration, pre-notification letters are sent to

all households selected for the survey. This letter informs households that they have been

selected for the survey and that they will be receiving a questionnaire in the mail. Following

this pre-notification letter, a questionnaire is sent out to each household in the sample. As

mentioned above, roughly 253,000 questionnaires are mailed out each month. If the household

does not return the questionnaire a reminder card is sent to the nonresponding household,

followed by another questionnaire if the household still does not respond.

Households that do not respond to the mail-in questionnaire and for whom the Census

Bureau has a telephone number are contacted for a telephone interview (a Computer-Assisted

Telephone Interview, CATI) roughly one month after the questionnaire was first mailed out. A

subsample of households (one out of every three) that do not respond to either the mail-in

questionnaire or the telephone interview are selected for a face-to-face interview (a Computer-

Assisted Personal Interview, CAPI) two months after the questionnaire was first mailed out.

The ACS questionnaire takes roughly 40 minutes to complete.

Unlike many other

surveys, including the NSFG and SIPP, there is no monetary incentive for completing the

questionnaire. With the mandatory nature of the ACS and high response rates, Census Bureau

staff do not currently see a need for monetary incentives.

Survey Modification. Most of the ACS content is mandated or required by law and

modifications to the survey require Office of Management and Budget (OMB) approval. The

Census Bureau does not have a vested interest in making changes to the content of the ACS

questionnaire. However, in recent years, Census staff worked with OMB and other Federal

agencies (e.g., DHHS) to identify and test changes to the questionnaire. The resulting changes

were incorporated into the 2008 questionnaire. As discussed above, the 2008 questionnaire

includes three new questions about marriage and divorce.

Much of the work required to change the 2008 ACS questionnaire occurred in 2005 and

2006. This included preparing preliminary question wording, testing the questions, and then

finalizing the wording. The 2008 questionnaire was finalized and approved by OMB in 2007.

OMB’s task is to ensure that the survey does not place undue burden on respondents. The 2008

questionnaire asks about change in marital status in the previous year, so no additional changes

are necessary for overall marriage and divorce rates to be calculated at the national, state and

local levels.

National Survey of Family Growth (NSFG)

The NSFG is a cross-sectional survey that provides retrospective information on marriage,

divorce, cohabitation, childbearing, and the health of women and infants. It is administered by

the National Center for Health Statistics.

NSFG data are currently available from six cycles

Information on survey administration is from U.S. Bureau of the Census. 2008a. Survey Basics: Data Collection. American

Community Survey. <http://www.census.gov/acs/www/SBasics/DataColl.htm>. Accessed January 2008.

This average is for the 2007 questionnaire.

There are no plans to change the marriage/divorce questions in the ACS beyond the changes introduced in the 2008 survey. The

Census bureau will, however, evaluate the quality of the responses to these questions to determine if any changes are necessary.

The National Center for Health Statistics is part of the Center for Disease Control and Prevention (CDC) and the CDC is part of

DHHS.

(1973, 1976, 1982, 1988, 1995, and 2002) and a seventh cycle is underway (2006-2010). The first

data from Cycle 7 is expected to be released in late 2009.

Unlike the ACS, the NSFG sampling unit is the individual, not the household. Over the

course of the NSFG cycles, the sample population has changed somewhat, although it has

always been limited to the civilian noninstitutionalized population between the ages of 15 and

44.

For the 1973 and 1976 surveys, the sample only included ever-married women, while the

1982, 1988, and 1995 surveys expanded to include all women (between 15 and 44).

For the 2002

and 2006–2010 surveys, the sample population includes both men and women (ages 15 to 44).

Because the survey excludes individuals over the age of 44, the marital histories are not

representative of the entire population (i.e., they do not represent marriage and divorce

experience after age 44).

The NSFG sample is significantly smaller than the ACS and SIPP samples. The sample

sizes for the 1995 and 2002 NSFG samples (Cycle 5 and 6) were 10,800 and 12,600, respectively.

Cycle 7, which is scheduled to be administered continuously from 2006 through 2010, is

expected to have a total sample size of 20,000, with an average of 5,000 interviews occurring per

year.

In essence, the Cycle 7 NSFG interviews are being spread over a series of years as a way

to increase the sample size without increased resources; cost savings result from the need to

train fewer interviewers, for example. Because the annual sample size is relatively small, the

first year of data will not be released alone. Rather, a data file containing between 12,000 and

13,000 interviews from 2006-2008 will be released in late 2009, with subsequent data files being

released about every 2 years. NSFG staff suggest that a minimum of two years of Cycle 7 data

be used to calculate marriage and divorce rates. Because the NSFG sample is relatively small, it

cannot be used to detect small changes in marriage and divorce rates over time.

The NSFG sample is representative only at the national and regional levels (four Census

regions) and cannot be used to calculate marriage and divorce rates at the state or local level.

The NSFG provides limited geographic information on the public use data file. Beginning with

the 2002 data, the public release files do not provide information on region (state of residence

was not on earlier files). Thus, researchers who want to examine marriage and divorce by

region of the country need to obtain access to the restricted data file.

There is only a single

geographic variable on the 2002 public release file—whether individuals live in a (1) Standard

Metropolitan Statistical Area (SMSA), central city; (2) SMSA, other; or (3) non-SMSA.

Marriage, Divorce, and Cohabitation Questions. In addition to current marital status,

the NSFG collects marital history information. Marital history for females is collected through a

series of questions that ask how many times the respondent has been married and details about

their marriages.

From 1973 to 1988, detailed information was collected on up to three

Unless otherwise noted, information is provided for the most recent data available—the 2002 NSFG.

Starting with the 2002 survey, someone who is active duty military and not living on a military base can be included in the

survey.

Prior to the 1988 survey, the sample was restricted to the continental United States.

The 2006-2010 sample population is somewhat broader, as it includes people living in group quarters.

While Cycle 7 of the NSFG is spread across five calendar years, it is considered a four-year cycle with interviews scheduled for

the four-year period from June 2006 through May 2010.

Each year of Cycle 7 interviews will be a nationally representative sample.

The restricted data file also includes information on state of residence and an indicator of urban versus rural.

See Appendix B for detailed questions.

marriages, and beginning in 1995, it was collected for all marriages. This information includes

the date of marriage, if and when the marriage ended, how the marriage ended (death, divorce,

or annulment), as well as information on separations.

Additionally, for each marriage, the

survey collects data on the husband, including whether or not he had been married before and

if he had any children from previous relationships. For the current husband (or first husband if

not currently married), the survey also asks about age, race, ethnicity, and educational

attainment. The male questionnaire collects largely the same marital history information as the

female questionnaire; however, it does not ask about separations within each marriage or about

the educational attainment of the former wife.

Current cohabitation status has been collected since 1973, but cohabitation histories were

first collected in 1995. Female respondents are asked, for each marriage, if they lived with their

spouse prior to marriage, the date they began cohabitating, and whether or not there were

periods when they stopped cohabiting. In addition, the NSFG collects information about

cohabiting partners the female respondent did not marry. Information is collected on the

current cohabiting partner and up to eight previous cohabiting partners. Collection of

cohabitation for males is much more limited; the survey asks about the dates of cohabitation

and demographic characteristics of only the first cohabiting partner.

Data Availability, Access, and Clarity. In Cycles 1 through 6, NSFG data have been

released to the public about two years after collection. Data from the 2002 survey was released

to the public in December 2004. The data release schedule for the next cycle of the NSFG is not

as straightforward, since the interviews will take place over several years—from 2006 through

2010. It is expected that the data collected from 2006 through 2008 will be released in late 2009,

with the final Cycle 7 data file being released about one year after data collection ends.

The NSFG public use data files are available on CD-ROM and can be requested free of

charge from NCHS. In addition to the raw data files, the CD-ROM includes the questionnaire

and data dictionary. For Cycle 6 (2002), the NSFG documentation is also on the internet with

interactive features, and therefore allows for updates as necessary.

Researchers wishing to use the restricted NSFG data can gain access through the NCHS

Research Data Center (RDC), located at the NCHS headquarters in Hyattsville, Maryland. One

must fill out an application to gain access to the data. NCHS’s goal is to ensure that the user

does not use the data illegally (i.e., attempt to identify individuals). Once access to the restricted

data has been granted, there are three ways to use the data. The first option (available for $200

per day) is to conduct the analysis on site at the RDC. This is costly because NCHS has a staff

member on call to answer questions and prevent inadvertent disclosures of confidential data.

The second, and cheapest, option ($500 per year) is to use the restricted data remotely with SAS;

users submit their SAS programs to the RDC via email, and output is emailed back to the user.

The third option is called “staff-assisted remote access,” in which a user submits a program and

an RDC staff member runs it. There is usually a per-run charge for this procedure. Most people

go to the RDC first for a day or two to get everything set up and then use it remotely

afterwards. However, because only SAS is used for remote access, people who wish to use

STATA or another program must do so in the RDC.

The NSFG provides information on both the month and year events (such as marriage and divorce) occur.

Information on educational attainment is collected for only the current wife.

Marriage, Divorce, and Cohabitation Correlates. The NSFG provides information on

many correlates of marriage, divorce, and cohabitation. A unique feature of the survey is a

battery of roughly 20 questions on attitudes toward marriage, parenthood, and gender roles,

which can shed light on patterns of marriage, divorce and cohabitation. In terms of the

individuals’ employment and economic situation, the survey asks individuals about

employment history, earnings, total income and income sources (of the individual and family),

and receipt of welfare benefits. The NSFG also provides data that enable researchers to examine

the relationship between individuals’ living arrangements throughout childhood and their

marriage/divorce decisions as adults. Family planning practices, pregnancy, birth, and

adoption histories are also collected. The survey collects selected information on the well-being

of respondents’ children. The female questionnaire includes questions about premature births,

birth weight, smoking during pregnancy, and breast-feeding. The male questionnaire asks

respondents about interactions and activities with children with whom he does and does not

live, such as helping with homework, eating meals together, and talking with them, as well as

how good of a father the respondent thinks he is. For children that the father does not live with,

the survey also asks how often the respondent visits his children and if he is satisfied with the

number of visits.

Survey Response Rate. The overall response rates were roughly the same in the two

most recent completed NSFG cycles—78.6 percent in the 1995 survey and 79 percent in the 2002

survey.

In the 2002 survey, the response rates were very similar for females and males (80

percent and 78 percent, respectively), but differed somewhat by race/ethnicity and age. The

highest response rates were for Hispanic and black teenagers (84 percent), and the lowest rates

were for adult white males ages 25 to 44 (approximately 75 percent). In the 1995 survey, the

response rates varied very little by race and ethnicity; all races and ethnicities had response

rates between 78 to 80 percent. However, response rates did differ significantly by age; women

ages 25 to 29 had a response rate of 75 percent, as opposed to the 83 percent rate for women

ages 15 to 17.

Survey Administration. Unlike the ACS, all NSFG interviews are conducted face-to-face

by the NSFG interviewer at the respondent’s home. Prior to the interview, an advanced letter

explaining the survey is sent to the household. NSFG interviews take place in-person because of

the sensitive nature of the questions as well as the complexity of the questionnaire (e.g.,

collecting dates and other detailed information on prior marriages, divorces, and cohabiting

relationships). These interviews are primarily conducted using Computer-Assisted Personal

Interviewing (CAPI). Because the survey includes questions of a sensitive nature (e.g., abortion

and sexual history), a portion of the NSFG is self-administered. This self-administered portion

The 1995 response rate is from Kelley, J. E., W. D. Mosher, A. P. Duffer, and S. H. Kinsey. 1997. “Plan and Operation of the 1995

National Survey of Family Growth.” Vital and Health Statistics, Series 1, No. 36. Hyattsville, MD: National Center for Health

Statistics. (See Table F and pages 82-89).

The 2002 response rate is from Groves, R. M., G. Benson, W. D. Mosher, J. Rosenbaum, P. Granda, W. Axinn, J. Lepkowski, and

A. Chandra. 2005. “Plan and Operation of Cycle 6 of the National Survey of Family Growth.” Vital and Health Statistics, Series

1, No. 42. Hyattsville, MD: National Center for Health Statistics. (See page 50).

Burstein, N., L. Lindberg, D. Fein, L. Page, and R. LaRock, Jr. 2003. “Guide to Data Sources on the Determinants of Marriage and

Cohabitation.” Bethesda, MD: Abt Associates. (See page 51).

of the interview is conducted using Audio Computer-Assisted Self-Interviewing (A-CASI), in

which respondents enter answers directly into the computer.

The 2006-2010 NSFG survey takes males an average of 60 minutes to complete and

females an average of 80 minutes. NCHS began offering incentives to respondents for their

participation in 1995. Respondents received $20 for completing the 1995 survey; the incentive

increased to $40 for the 2002 and 2006-2010 surveys.

Survey Modification. Modifications to the NSFG can be suggested by current or

potential funding agencies (funding agencies are also referred to as “collaborators”). There are

three major collaborators—NCHS, National Institute of Child Health and Human Development

(NICHD), and the Office of Population Affairs at DHHS; other collaborators include Division of

Reproductive Health (within the Centers for Disease Control and Prevention, CDC), Divisions

of HIV/AIDS Prevention (within the CDC), Office of the Assistant Secretary for Planning and

Evaluation (ASPE, within DHHS), Administration for Children and Families (ACF, within

DHHS), among others. Proposals to modify the survey should be specific and be submitted to

NSFG staff at NCHS. For each cycle of the NSFG, the collaborators meet to design a survey

instrument that meets their needs. Proposed changes to the NSFG are discussed by the

collaborators, with meetings of the collaborators occurring on an as needed basis. For the Cycle

7 survey, which began in July 2006, the questionnaire was finalized by June 2005.

Each cycle of the NSFG requires Institutional Review Board (IRB) and OMB approval.

The IRB process is an internal process within NCHS that seeks to guarantee that survey

participation is informed and voluntary, and that the interests of respondents are protected.

As noted above, the OMB process is designed to ensure that the survey does not place undue

burden on respondents and that it has practical utility for federal government programs. Prior

to OMB submission, the NSFG survey package must be reviewed and approved by NCHS,

CDC, and ASPE (in that order). Completing the IRB and OMB reviews takes six to eight

months. After the survey has been submitted for OMB and IRB approval, it is still possible to

modify the survey. The difficulty involved in modifying the survey depends on the extent of the

proposed changes and the degree to which new questions elicit sensitive, personal information.

The NSFG provides national estimates of marriage and divorce rates, but does not

provide state and local level estimates. Given recent changes to the ACS questionnaire (in 2008),

along with the fact that the NSFG sample size and survey budget would have to increase

substantially to provide state- or local-level estimates of marriage and divorce rates, the NSFG

is not an appropriate vehicle for obtaining this information. All of the NSFG interviews take

place face-to-face so are relatively costly to administer.

One improvement that would make the sample more representative of marriages and

divorces in the U.S. would be to expand the age range of the NSFG beyond individuals ages 15

to 44. In 1995, 13 percent of all men and 9 percent of all women who married were over age 44.

After the self-administered portion of the interview is complete, the respondent’s responses are locked and cannot be accessed

until the data are processed at NCHS survey headquarters.

The IRB process is administered by the CDC.

Authors’ tabulations from the 1995 Marriage and Divorce Data of the National Vital Statistics System of the National Center for

Health Statistics. 1995 is the most recently available vital statistics data.

Similarly, 24 percent of all men and 13 percent of all women who divorced were over age 44.

A broader age range would provide substantially more information on marriage dissolution

and remarriage. An expansion of the age range should be accompanied by an increase in

sample size. The NSFG is used to study demographic events, like childbearing, that tend to

occur among younger adults (ages 15 to 44). Thus, it is important to maintain the number of 15-

to 44-year-olds who are surveyed.

Obtaining information on those over age 44 should be

accomplished by adding additional individuals to the sample.

While it would not make the

rate determined through the NSFG representative of the entire U.S. population, as a source of

cost-savings, the NSFG could limit the expansion of the sample to individuals ages 45 to 54. Age

54 is a reasonable upper bound because vital statistics data suggests that only a small fraction of

marriages and divorces occur to individuals over age 54. In 1995, 5 percent of all men and 3

percent of all women who married were over age 54 and 7 percent of all men and 4 percent of

all women who divorced were over age 54.

Another potential NSFG improvement would be to add marriage and relationship

quality questions to the questionnaire. This important correlate of marriage and divorce is often

omitted from large national data sets, including the NSFG (and ACS and SIPP). Incorporating

these questions into the NSFG would allow researchers and policymakers to better understand

the relationship between marital status changes, marital and relationship quality, and

demographic and behavioral characteristics. While the NSFG has never asked questions about

marital quality, the format of the NSFG interviews makes it a viable option for collecting these

data. The NSFG interviews are face-to-face and also have a self-administered component, so

respondents may feel more comfortable answering these questions honestly than in other

survey formats.

Survey of Income and Program Participation (SIPP)

The SIPP is a panel study that focuses on the income and program participation of

households and individuals, and is administered by the U.S. Census Bureau. Panels began

annually from 1984 to 1993 and in 1996, 2001, and 2004. In addition, a 2008 SIPP panel began in

September 2008. Each panel generally lasts between two and four years, with more recent

panels lasting three to four years. Respondents are interviewed every four months about the

previous four months, a period referred to as a wave. Each SIPP panel consists of a core

questionnaire, which is administered in each wave, and topical modules, which collect

supplemental information on a variety of topics. The number of topical modules and the topics

that they cover vary by wave. All civilian non-institutionalized households are eligible to be

sampled. The sample size has varied across panels; the 1996 panel includes 36,800 households,

the 2001 panel 35,100 households, and the 2004 panel 46,500 households.

Authors’ tabulations from the 1995 Marriage and Divorce Data of the National Vital Statistics System of the National Center for

Health Statistics.

This point was emphasized by NCHS staff.

At this time, NCHS and the NSFG's other funding agencies do not have the funds to expand the age range.

Author’s tabulations from the 1995 Marriage and Divorce Data of the National Vital Statistics System of the National Center for

Health Statistics.

Adding relationship quality questions to the NSFG may have monetary and/or non-monetary costs. Lengthening the survey

would add monetary costs, while dropping questions to make room for the new material would mean giving up the

information that could have been garnered from the deleted material. However, if only a small number of questions (e.g., 1-5)

were added, then the cost could be modest.

The SIPP panels have typically provided representative estimates at the national and

regional levels.

The 2004 panel also provides representative estimates for about 30 states

(earlier panels did not provide a representative sample for any state). However, it will be

difficult to detect whether small differences across states are statistically significant, because of

the relatively small sample size per state. In addition to providing information on state of

residence,

the SIPP identifies large metropolitan areas (MSAs), although the SIPP samples are

not representative of the populations in the MSAs. The public use data file also has a

metropolitan/residual residence variable. However, for confidentiality purposes, some

respondents who live in metropolitan areas of states with small non-metropolitan areas are

recoded as living in non-metropolitan areas.

Marriage, Divorce, and Cohabitation Questions. Marriage and divorce data are

collected for all individuals in the surveyed household who are age 15 or older. The most

reliable data from which to calculate marriage and divorce rates come from the marital history

topical module, which is administered in the second wave of each panel.

This topical module

asks individuals the number of times married, dates of marriage, dates of separation, dates that

the marriage ended, and why it ended (e.g., death or divorce). The module covers up to three

marriages—the first, second, and most recent. Because this retrospective history is collected

early in the survey, attrition-related issues that hamper longitudinal event data are not likely to

be problematic. The survey collects the month and year of marriages and divorces (as well as

the month and year of births), but starting with the 2001 panel, the public release data

suppresses the month of these events to ensure confidentiality. As a result, one cannot use the

SIPP to study marriage duration where duration is measured in months.

SIPP collects information on current cohabitation status, but does not collect

cohabitation histories. Information on current cohabitation is collected through a “relationship

to reference person” question on the core questionnaire. Thus, for the reference person,

cohabitation status is collected for each wave of the panel. Cohabitation status for other

household members is available once during the panel—the Wave 2 household relationships

topical module identifies the relationship of each individual in the household to one another.

Census has tested questions that directly ask about cohabitation status, but not cohabitation

history.

Data Availability, Access, and Clarity. There is a relatively long lag time between SIPP

data collection and data availability. In the past, Wave 1 core data typically have been released

approximately 20 months after data collection. For the 2004 panel, the Census Bureau released

the preliminary version of the Wave 1 core data in April 2006, and the final version was

released in January 2008. Final versions of the core data for Waves 2 through 4 were also

released in January 2008. As of September 2008, core data were available through Wave 9. The

Calculating marriage and divorce rates for all four Census of regions is not possible with the public release SIPP data prior to

the 2004 panel. The public release file did not identify region of the country, and the variable that identified state of residence

grouped together states that cross regional boundaries. In the 2001 panel, for example, North Dakota, South Dakota, and

Wyoming were combined into a single group—North Dakota and South Dakota are in the Midwest and Wyoming is in the

West.

In the 1996 and 2001 panels, Maine and Vermont were combined into a single group, as were North Dakota, South Dakota, and

Wyoming. In 2004, however, each state is identified individually.

See Appendix B for detailed questions.

Census Bureau also began releasing topical module data in April 2006. Subsequent waves of

core and topical module data are scheduled to be released by January 2009.

Data can be downloaded for free from the DataFerrett application and from the SIPP

website. The user can also download SIPP documentation, such as questionnaires and

codebooks. The Census Bureau provides many SIPP working papers on topics including the

accuracy of SIPP data and using SIPP data for policy analysis.

The SIPP data can be difficult to

use because of its longitudinal design. However, using only the Wave 2 data (the wave with

marital history information) is relatively straightforward; the user simply has to merge together

the core and the topical module data.

Users can gain access to the restricted use data, which are located at the Census Bureau's

Research Data Centers (RDC).

To gain access to the restricted data, users must a submit

proposal to the Census Bureau’s RDC and the Census Bureau's Center of Economic Studies.

These proposals are reviewed on several criteria; most importantly, the proposals must

demonstrate that the research will benefit Census programs.

If approved, researchers must

apply for Special Sworn Status, which includes fingerprinting and a background check. Users

should expect to wait at least six months between submitting the final proposal and beginning

research.

Marriage, Divorce, and Cohabitation Correlates. The SIPP provides a wide variety of

correlates beyond demographic characteristics. The core questionnaire collects information

about household composition, employment (e.g., hours worked per week, weeks worked, wage

rate, occupation), income (e.g., earned and unearned income, income sources, and investments),

and program participation. Topical modules supplement the core questionnaire on a range of

issues, including benefit recipiency history, fertility history, child care arrangements, child

support agreements, child well-being, and adult well-being. The child well-being topical

module, which was administered in Waves 3 and 9 of the 2004 panel asks questions about early

childhood experiences, such as child care arrangements; parent-child interactions, including

meals eaten and outings together; school-age enrichment, such as extracurricular activities;

child’s academic experience; parental educational expectations; and the parent’s feelings toward

his/her children. The adult well-being topical module, administered in Waves 5 and 11 of the

2004 panel, addresses living conditions, basic needs, and adequacy of food.

Survey Response Rate. Reporting the survey response rate for the SIPP is not as

straightforward as for the ACS and NSFG, because response rates differ over the panel due to

attrition. For Wave 1 (the initial interview), the response rate for the 2004 panel was 85.1

percent, slightly lower than the Wave 1 response rate of 86.7 percent for the 2001 panel. Recall

Vaughn, D.R. 2000. “Type of OASDI Benefit and Year of Death based on an Exact Match to Social Security Administration

Benefit Records, 1990 and 1991 Panels of the Survey of Income and Program Participation (SIPP): Description of the

Development of the Data for Public Release and a Preliminary Evaluation of Data Quality.” Washington, DC: U.S. Bureau of the

Census. Weinberg, D.H. 2002. “The Survey of Income and Program Participation—Recent History and Future Developments.”

Washington, DC: U.S. Bureau of the Census.

There are nine Census Bureau RDCs, located in Washington, DC, Cambridge, MA, Los Angeles, CA, Berkeley, CA, Durham,

NC, Ann Arbor, MI, Chicago, IL, New York, NY, and Ithaca, NY.

There are several ways to satisfy this requirement, including improving the quality of the data, leading to improved

methodology, improving imputations, developing linkages across time or surveys, or improving the sampling frame.

For a more complete overview of the process, see U.S. Bureau of the Census, The Center for Economic Studies. 2008b. The

Research Data Program. < http://www.ces.census.gov/index.php/ces/researchguidelines >. Accessed January 2008.

that the response rate is roughly 97 percent for the ACS and 80 percent for the NSFG. The

response rate in the second wave is important, because marital history is collected in Wave 2.

The Wave 2 response rate was 78.1 percent for the 2004 panel and 79.1 percent for the 2001

panel.

Survey Administration. The SIPP uses both face-to-face and telephone interviews,

which are conducted using Computer-Assisted Interviewing (both CAPI and CATI). Initial

interviews are always conducted in person, as are interviews with households that have moved.

Use of telephone interviews has varied across panels. The 2001 panel also conducted face-to

face interviews for Wave 2 and used telephone interviews when possible for Wave 3 through

Wave 9, while the 2004 panel maximizes the use of telephone interviews beginning with Wave

The survey respondent (i.e., the SIPP reference person) for the household is the person

in whose name the residence is owned or rented. For households owned jointly by a married

couple, either spouse can be the reference person.

The SIPP attempts to interview all

individuals 15 or older in each sampled household. If individuals are unavailable or

uncooperative, proxy interviews (usually from the reference person) are accepted. Census staff

estimates that approximately 25 percent of interviews are proxy interviews. This is a drawback

of the SIPP because proxy respondents may be less familiar with individuals’ experiences, such

as their marital histories.

There are monetary incentives for some SIPP respondents. For households that are

uncooperative, interviewers can offer a $40 debit card as an incentive to participate. On average,

households take 42 minutes to complete the core questionnaire and 21 minutes to complete the

topical modules; thus, each wave takes households an average of 63 minutes to complete.

However, the length of the survey varies by wave, as waves have different numbers of topical

modules administered.

Survey Modification. To propose a change to the SIPP survey, the interested agency

should contact Census Bureau staff. The Census Bureau is organized around topic areas (not

data sets) and proposals related to marriage and divorce should be addressed to the chief of the

Population Division.

The Census Bureau evaluates proposals along several dimensions,

including implications for respondent burden, administration of questions (e.g., where new

questions will fit into the survey), sensitivity of questions, and the need for cognitive testing. If

necessary, the Center for Survey Methods Research within the Census Bureau tests new and

revised questions. The SIPP is funded by the Census Bureau, so agencies that propose a change

to the SIPP are not responsible for providing the monetary resources needed to implement the

change. However, the background materials, technical support, and data evaluation are the

agencies’ responsibility.

The response rate generally falls over the panel. The overall response rate for the 2001 SIPP panel was 68.1 percent, for example.

In the first wave of the 2001 panel, the wife was the reference person in 34 percent of married households.

These numbers are calculated based on an average of 2.1 adults age 15 years and older per household.

Requests related to the TANF or food stamp programs, for example, would be addressed to the Division of Housing and

Household Economic Statistics.

Any changes to the SIPP must be agreed upon by the OMB Interagency SIPP Committee,

which was established in 1983 to assist the Census Bureau in developing and revising the SIPP

questionnaire.

In addition to OMB and Census staff, the Committee comprises individuals

from many agencies including DHHS, Department of Labor, Department of Agriculture, and

Department of Housing and Urban Development. There are no regularly scheduled meetings.

Once the questionnaire is complete and has received Census Bureau approval, the

survey, along with complete documentation, is sent to OMB for approval. Census staff must

prepare a separate OMB package for each wave. The Wave 1 OMB package includes the core

questionnaire and the Wave 1 topical modules. Since the core questionnaire does not change

across waves, OMB packages for subsequent waves only need to include topical module

questionnaires for approval. Since the different topical modules represent different burdens,

approval for each wave is necessary. The OMB clearance process takes nine months on average.

Even after questions have been approved and appeared on the SIPP, the data garnered

from them may not become available. Census reviews the questions and responses under its

“disclosure risk process” to ensure the confidentiality of respondents. This process occurs

approximately 60–90 days before each public data file release.

The process of making modifications to the SIPP questionnaire needs to begin at least

two years prior to the start of the panel. The entire survey is locked at least one year before

interviewing begins, depending on cognitive testing, instrument development, and processing

considerations.

It would not be cost effective, given the recent changes to the ACS questionnaire, to

expand the SIPP sample to provide state-level marriage and divorce rates. One improvement to

the SIPP’s usefulness in studying marriage and divorce would be to add questions about

relationship quality. The Census Bureau's Survey of Program Dynamics (SPD), which was a

continuation of the 1992 and 1993 SIPP panels, included three questions on relationship quality

and conflict that appeared on the self-administered section of the questionnaire. Since the SPD

questions were previously approved by the Census Bureau, they would not need cognitive

testing. The SPD asked questions about (1) how happy the individual was with his/her

relationship, (2) how often the couple discussed or considered separating from one another, and

(3) whether arguments between partners had become physical in the last year. This relationship

quality and conflict information, however, was never released to the public because of

confidentiality concerns. According to Census Bureau staff, the questions were determined to be

too sensitive in nature to release respondents’ answers to the public. Whether relationship

quality questions could be released if asked in future panels of the SIPP would need to be

reviewed by the Census Bureau.

Summary

All three data sets described above offer current national estimates on marriage and

divorce and collect data on an ongoing basis. However, they have weaknesses along with their

strengths.

Citro C.F. and G. Kalton (eds.). The Future of the Survey of Income and Program Participation. Washington, D.C.: National

Academy Press, 1993: 227-244.

• The ACS is best suited to provide state- and local-level information on marriage and

divorce because its sample size will enable state-level estimates of marital status on an

annual basis and county-level statistics based on three to five years of pooled data.

Beginning with the 2008 ACS, changes in marital status can be identified, making it an

ideal source for estimating local-level marriage and divorce rates.

• The NSFG provides the most detailed information on relationship histories and is the

only survey of the three that collects a cohabitation history. With the change to

continuous interviewing in 2006, the NSFG will provide updated national estimates of

marriage on an annual basis.

However, the NSFG’s relatively small sample size means

that it cannot be used to detect small changes in marriage and divorce rates over time or

provide state- and local-level estimates. Finally, as currently designed, NSFG

respondents are limited to individuals between the ages of 15 and 44. Its rich data about

marital, sexual and family history makes the NSFG an important data source for

understanding the correlates of changes in marriage and divorce rates.

• The SIPP collects marital histories that can be used to calculate marriage and divorce

rates at the national level and for the four Census regions. The 2004 SIPP panel provides

a representative sample for 30 states, but Census staff caution against using the SIPP to

calculate state-level marriage and divorce rates because the state sample sizes are too

small to accurately estimate the extent to which these relatively rare events occur.

Additionally, because of the panel structure of the SIPP, it cannot provide annual

marriage and divorce rates. Rather, it provides reliable measures of marriage and

divorce once every panel—or once every three to five years. However, like the NSFG,

the rich set of data collected by the SIPP, especially around income and program

participation, makes it an important data source for the understanding of changes in

marriage and divorce rates.

Recall that NCHS staff recommended that marriage and divorce rates be calculated with a minimum of two years of data (from

the Cycle 7 interviews that will run from 2006 through 2010).

V. REFERENCES

Becketti, S., W. Gould, L. Lillard, and F. Welch. 1988. “The Panel Study of Income Dynamics

after Fourteen Years: An Evaluation.” Journal of Labor Economics 6(4): 472–492.

Bramlett, M.D., and W. D. Mosher. 2002. “Cohabitation, Marriage, Divorce, and Remarriage in

the United States.” National Center for Health Statistics. Vital Health Statistics 23(22).

Bumpass, L., K. Raley. 2007. “Measuring Separation and Divorce.” In Handbook of Measurement

Issues in Family Research, edited by Sandra L. Hofferth and Lynne M. Casper (125–44).

Mahwah, NJ: Lawrence Erlbaum Associates.

Burstein, N., L. Lindberg, D. Fein, L. Page, and R. LaRock, Jr. 2003. “Guide to Data Sources on

the Determinants of Marriage and Cohabitation.” Bethesda, MD: Abt Associates.

Citro, C.F., and G. Kalton, eds. 1993. The Future of the Survey of Income and Program Participation

(227-44). Washington, DC: National Academy Press.

Clarke, S.C. Advance Report of Final Marriage Statistics, 1989 and 1990. Monthly Vital Statistics

Report 1995; 43(12): 7.

Fitzgerald, J., P. Gottschalk, and R. Moffitt. 1998. “An Analysis of Sample Attrition in Panel

Data: The Michigan Panel Study of Income Dynamics.” Journal of Human Resources 33(2):

251–99.

Groves, R. M., G. Benson, W. D. Mosher, J. Rosenbaum, P. Granda, W. Axinn, J. Lepkowski, and

A. Chandra. 2005. “Plan and Operation of Cycle 6 of the National Survey of Family

Growth.” Vital and Health Statistics, Series 1, No. 42. Hyattsville, MD: National Center

for Health Statistics.

Hill, D. H. 1997. “Adjusting for Attrition in Event-History Analysis.” Sociological Methodology 27:

393–416.

Kelley, J. E., W. D. Mosher, A. P. Duffer, and S. H. Kinsey. 1997. “Plan and Operation of the 1995

National Survey of Family Growth.” Vital and Health Statistics, Series 1, No. 36.

Hyattsville, MD: National Center for Health Statistics.

Kreider, R. M., and J. M. Fields. 2002. “Number, Timing, and Duration of Marriages and

Divorces: 1996.” Current Population Reports, Series P-70, No. 80. Washington, DC: U.S.

Bureau of the Census.

Lewin Group. 2008. “Collection of Marriage and Divorce Statistics by States.” Draft report to the

U.S. Department of Health and Human Services.

Lillard, L. A., and C. W. Panis. 1998. “Panel Attrition from the Panel Study of Income Dynamics:

Household Income, Marital Status, and Mortality.” Journal of Human Resources 33(2):

437–57.

Lillard, L.A., and L. J. Waite. 1989. “Panel versus Retrospective Data on Marital Histories:

Lessons from the PSID.” In Individuals and Families in Transition: Understanding Change

through Longitudinal Data (243-53). Washington, DC: U.S. Bureau of the Census.

Manning W., and P. Smock. 2003. “Measuring and Modeling Cohabitation: New Perspective

from Qualitative Data.” Center for Family and Demographic Research Working Paper

#2003-10.

U.S. Bureau of the Census. 2001. SIPP User’

s Guide, Third Edition. Washington, DC: U.S. Bureau

of the Census.

U.S. Bureau of the Census. 2002. “Meeting 21

Century Demographic Data Needs—

Implementing the American Community Survey: May 2002. Report 2: Demonstrating

Survey Quality.” Washington, DC: U.S. Bureau of the Census.

U.S. Bureau of the Census. 2008a. “Survey Basics: Data Collection. American Community

Survey. Washington, DC: U.S. Bureau of the Census.

http://www.census.gov/acs/www/SBasics/DataColl.htm.

U.S. Bureau of the Census, Center for Economic Studies. 2008b. The Research Data Program.

Washington, DC: U.S. Bureau of the Census.

http://www.ces.census.gov/index.php/ces/researchguidelines.

Vaughn, D.R. 2000. “Type of OASDI Benefit and Year of Death based on an Exact Match to

Social Security Administration Bene

fit Records, 1990 and 1991 Panels of the Survey of

Income and Program Participation (SIPP): Description of the Development of the Data

for Public Release and a Preliminary Evaluation of Data Quality.” Washington, DC: U.S.

Bureau of the Census.

Weinberg, D.H. 2002. “The Survey of Income and Program Participation—Recent History and

Future Developments.” Washington, DC: U.S. Bureau of the Census.

Zabel, J.E. 1998. “An Analysis of Attrition in the Panel Study of Income Dynamics and the

Survey of Income and Program Participation with an Application to a Model of Labor

Market Behavior.” Journal of Human Resources 33(2): 479–506.

VI. APPENDIX A: OTHER DATA SETS EXAMINED

In addition to the American Community Survey (ACS), the National Survey of Family Growth

(NSFG), and the Survey of Income and Program Participation (SIPP), 17 other data sets were

examined for their ability to measure marriage and divorce. However, based on the criteria

discussed in Section II, it was concluded that these data sets were unable to produce valuable

marriage and divorce statistics. In fact, as Exhibit A.1 shows, most of these data sets failed to

meet more than one of the criteria.

The following section describes each of these data sets (in

alphabetical order) and how they failed to meet the criteria. Much of the information regarding

these data sets is current as of January 2005.

Exhibit A.2 summarizes the survey

characteristics.

Current Population Survey. The Current Population Survey (CPS) is a monthly cross-

sectional survey that includes approximately 57,000 households and 112,000 individuals. The

survey captures the civilian, non-institutionalized population age 15 and older. These data

provide representative estimates on a national level, and data pooled from multiple years can

yield state-level estimates. The CPS provides information on marital status, but does not

provide information to calculate marriage and divorce rates. The CPS did include a marital

history supplement roughly every five years from 1971 to 1995, which enabled calculation of

marriage and divorce rates, but it was discontinued. According to Census Bureau staff, the

Wave 2 topical module of the SIPP has essentially replaced the CPS marital history supplement,

and the Census Bureau currently has no plans to reintroduce it. Prior to 2007, cohabitation

status was obtained from a question about "relationship to household respondent," so was only

available for survey respondents and their partners. However, beginning in 2007, the CPS

began to ask a direct question about cohabitation, so now provides this information for all adult

household members.

Decennial Census. The Decennial Census has traditionally collected information on

marital status, but the extensiveness of the information collected has changed over time. In the

2000 Decennial Census, current marital status was collected only on the long form. In 1990, it

was collected on both the long and short forms.

Because the ACS is replacing the Decennial

Census long form, marital status is not expected to be collected in the Decennial Census starting

in 2010 and going forward. The Decennial Census currently provides information on the

cohabitation status of household respondents and their partners. Like the ACS, cohabitation

status is obtained from a question that asks about relationship to household respondent. This

information is collected on the short form and is expected to continue in future rounds of the

Decennial Census.

Because each of these data sets has failed to meet at least one of the primary criteria, we do not discuss the availability of

correlates of interest, which is a secondary criterion.

The information reported in this appendix is from the most current survey as of January 2005, except for selected updates where

information was readily available.

The Decennial Census long form has traditionally been given to five percent of the U.S. population and asks more detailed

questions than the Decennial Census short form questions, whereas the Census Bureau aims to have all individuals represented

in the short form questionnaire.

Exhibit A.1: Evaluation Criteria Attained by Other Data Sets

Name of Survey Data Set

Representativeness

Relevancy Reliability

Broad Population State/Local

Ongoing

Current Population Survey (CPS)

x x

Decennial Census x x x

Early Childhood Longitudinal Study

- Birth Cohort (ECLS - B)

x x

Early Childhood Longitudinal Study

- Kindergarten Cohort (ECLS - K) x x

Fragile Families and Child Well-

Being Study x x

Health and Retirement Study (HRS) x x

Marital Instability Over the Life

Course Study x

Medicare Current Beneficiary

Survey (MCBS)

x x

National Health Interview Survey

(NHIS)

x x

National Longitudinal Survey of

Mature Women (NLSMW) x

National Longitudinal Survey of

Young Women (NLSYW)

National Longitudinal Survey of

Youth - 1979 (NLSY79) x x

National Longitudinal Survey of

Youth - 1997 (NLSY97)

x x

National Survey of America's

Families (NSAF) x x x x

National Survey of Families and

Households (NSFH)

Panel Study of Income Dynamics

(PSID)

x x

Survey of Program Dynamics

(SPD) x

* An x indicates that the data set fulfills the criterion.

All information is current as of January 2005, except for selected updates where information was readily available.

For a data set to meet this criterion it must be able to provide recent estimates of marriage and divorce rates. Data sets that do

not meet the ‘relevancy’ criterion (i.e., cannot be used to compute marriage and divorce rates) will not meet the ‘reliability’

criterion (i.e., marriage and divorce rates likely to match an external source).

The Current Population Survey included a supplement gathering retrospective marital histories approximately every five years

from 1971 to 1995. It has since discontinued this supplement, so it no longer meets the ‘relevancy’ and ‘reliability’ criteria.

The CPS can provide state-level estimates by pooling two or more years of data.

The CPS still collects marital status at a point in time, but no longer collects marital histories.

The Decennial Census is ongoing, but starting in 2010, it will no longer collect marital status information.

The National Survey of America's Families has representative estimates for 13 states.

Early Childhood Longitudinal Studies: ECLS-K and ECLS-B. The Early Childhood

Longitudinal Studies began in 1998 for the kindergarten cohort (ECLS-K) and in 2001 for the

birth cohort (ECLS-B). Each study consists of a baseline interview (at entrance to kindergarten

and at birth, respectively) with several follow-up interviews. It collects data for approximately

13,500 to 22,000 children and their parents. The survey collects information on parents’ marital

and cohabitation status at each interview date, enabling the study of the effects on marriage,

divorce, and cohabitation on children. However, this information is only collected

longitudinally (at each interview), so the reliability of the data is an issue for these two data sets.

Additionally, the sample for adults is only representative (on the national level) of the parents

with children in those two cohorts.

Fragile Families and Child Wellbeing Study. The Fragile Families and Child Wellbeing

Study is a longitudinal survey of the mothers and fathers of newly born children in large city

hospitals, with an emphasis on parents with non-marital births. Baseline interviews with 4,700

families took place between 1998 and 2000, with follow-up interviews one, three, and five years

later. The baseline data and the first and third year follow-up data are available. The fifth year

follow-up data is expected to be available in the summer/fall 2008. The survey follows the

relationship quality between the mother and father, regardless of marital status, as well as any

other relationships the parents may have with other partners. However, this data is collected

longitudinally and represents only a very small, distinct subpopulation.

Health and Retirement Study. The Health and Retirement Study (HRS) is an ongoing

panel survey of older Americans that provides retrospective marriage and divorce events

information. The HRS began in 1992 and interviews individuals biennially. The sample includes

individuals born in 1953 or earlier (ages 51 and older in 2004) and has a total of five cohorts. The

two original cohorts, individuals born between 1931 and 1941 and individuals born in or prior

to 1923, were first sampled in 1992 and 1993. Two other cohorts were added in 1998,

and a

fifth cohort was added in 2004. The HRS collected marital histories in the initial wave of the

panel, for up to three marriages. For marital status changes since the initial wave, data users

must rely on the longitudinal data. Additionally, since the HRS collects data only on older

Americans, it captures only a small portion of marriages and divorces and is not representative

for the population at large.

Marital Instability Over the Life Course Study. The Marital Instability Over the Life

Course Study is a panel survey that began with 2,000 households in 1980 and continued

through 2000. The sample includes married individuals who were between the ages of 18 and

55 in 1980. While the main focus of the study is marriage quality, it does collect a retrospective

marital history. However, these estimates are representative only of married individuals in

1980, which makes studying marriage and divorce with this data set problematic. Moreover, the

small sample only allows for analysis at the national level. Finally, the panel ended in 2000 and

there are no plans to start a new panel.

Medicare Current Beneficiary Survey. The Medicare Current Beneficiary Survey

(MCBS) is an ongoing survey of 12,000 individuals from the Medicare enrollment file, with data

currently available from 1991 to 2002. Each individual is part of the study for four years, with

The cohort introduced in 1993 was originally a separate study, the Study of Assets and Health Dynamics Among the Oldest Old

(AHEAD). The two studies were merged in 1998.

interviews taking place three times a year. Panels are rotated out of the survey each year, and

new panels are added to replace old ones. The main focus of the MCBS is on health status,

insurance coverage, access to care, and medical costs, but the study does collect data on marital

status. The marital status data are collected longitudinally (not retrospectively), so the data on

changes in marital status might not be wholly reliable. Also, since the sample population is

drawn from Medicare beneficiaries, it is representative of only the subpopulation of older

Americans on Medicare.

National Health Interview Survey. The National Health Interview Study (NHIS) is a

cross-sectional survey of 36,000 households from 1957 to the present. Its main focus is on

individuals’ health problems, health insurance, access to coverage, and behavior that may lead

to future health issues. In addition, the questionnaire asks about marital status, with

cohabitation embedded within the marital status question. However, it does not ask about

changes in marital status, so marriage and divorce rates cannot be calculated.

National Longitudinal Studies: NLSMW, NLSYW, NLSY79, and NLSY97. The

National Longitudinal Surveys (NLS), a project of the U.S. Bureau of Labor Statistics, track the

lives of selected respondents for upwards of 30 years. There are four National Longitudinal

Survey data sets that we consider—NLS Mature Women (NLSMW), NLS Young Women

(NLSYW), NLS Youth cohort 1979 (NLSY79), and NLS Youth cohort 1997 (NLSY97).

Initially,

data for each of the samples were collected annually; however, the NLSY97 cohort is the only

group still followed each year.

These surveys collect data on current marital status and

changes in marital status since the last interview. Additionally, in recent years, cohabitation

status and changes in cohabitation status has also been collected. While the NLS has detailed

marriage and divorce data for the past 30 years for several subgroups, problems of attrition lead

to questions of its reliability. Additionally, these studies can only provide estimates at the

national level and not at state or local levels.

National Survey of America’s Families. The National Survey of America’s Families

(NSAF) is a survey of 43,000 households administered in 1997, 1999, and 2002. The survey is

representative of the civilian, non-institutionalized population under 65 for the nation as a

whole as well as thirteen oversampled states. The NSAF provides detailed relationship

information of all household members, thereby capturing cohabitation status for all household

members. The 1997 and 1999 surveys only collected current marital status, although the 2002

survey collected information on change in marital status, making it possible to calculate

marriage and divorce rates. However, there are no plans to field another survey, so the NSAF

cannot be use to measure marriage and divorce in the future.

National Survey of Families and Households. The National Survey of Families and

Households (NSFH) is a panel survey of 9,600 households, sampling non-institutionalized

civilians aged 19 and older. The survey included three waves and was administered in 1987-88,

1992-94, and 2001-02. There are no plans for additional waves of the survey. The NSFH

More specifically, these four cohorts are the 1968 Mature Women cohort (ages 30–44 as of 3/31/67), 1968 Young Women cohort

(ages 14–24 as of 3/31/66), 1979 Youth cohort (ages 14–24 as of 12/31/78), and 1997 Youth cohort (ages 12–16 as of 12/31/96).

Data for the mature women and young women cohorts were last collected in 2003; no future collection is planned. In addition to

these cohorts, there were two additional male cohorts that have since been discontinued. These are the 1967 Older Men cohort

(ages 45–59 as of 3/31/66) and the 1967 Younger Men cohort (ages 14–24 as of 3/31/66).

The NLSY79 was fielded annually until 1994 and biennially since that time.

collected retrospective marital and cohabitation histories as part of the first wave and has

collected marital status and changes in marital status for each subsequent wave. In addition, the

survey collects information on marriage quality, including amount of time spent together and

how they resolve disputes. The NSFH sample was representative of the U.S. population in the

late 1980s, but is not representative of today's population. Further, the sample does not permit

representative estimates at the state or local levels.

Panel Survey of Income Dynamics. The Panel Survey of Income Dynamics (PSID) is an

ongoing panel survey of roughly 8,000 families and 37,500 individuals that has been running

since 1968. Respondents were interviewed annually between 1968 and 1996 and biennially since

1997. While the PSID moved from annual to biennial collection in 1997, the survey directly asks

questions about marital status changes since the last survey, capturing all marriage disruptions.

The PSID also collected retrospective marital histories in 1985, 17 years after the start of the

survey. Cohabitation status is gathered for the household head only. Because of the longitudinal

nature of the PSID and the fact that many years elapsed between the start of the survey and the

collection of marital histories, there are concerns that marriage and divorces statistics from the

PSID are not reliable due to attrition. The PSID is also not representative of the current U.S.

population, since the main sample was drawn in 1968.

Survey of Program Dynamics. The Survey of Program Dynamics (SPD) is a

continuation of the 1992 and 1993 SIPP panels yielding data for roughly 16,000 households from

1992 until 2002. Retrospective marriage history data in the SPD is from the SIPP panel. Beyond

what was collected in the SIPP, the SPD collected marital status information in each wave. In

addition, the SPD included questions about marriage quality, such as how happy respondents

are with their relationships, how often and why conflicts occur with their spouse/partner, and

whether they have considered separating. This relationship quality and conflict information,

however, was never released to the public because of confidentiality concerns. While the SPD

collects data on marriage and divorce, there are issues with the SPD that preclude it from being

used to measure marriage and divorce rates. The first problem is that the SPD has had

significant amounts of attrition over the course of the panel. Secondly, there are no plans to

continue the SPD in the future. Finally, the sample is only large enough to produce

representative estimates at the national level and not the state level.

The PSID introduced a sample of immigrants (with a focus on Latinos) in 1990, but dropped them in 1995.

Exhibit A.2: Summary of Other Data Sets

Name

Current Population

Survey (CPS)

Population

Non-institutionalized

civilians ages 15 and over

Sample Size

57,000 HH

(112,000

individuals)

•

Survey Design

Cross-sectional

Monthly data

Available: 1945 - 2007

Nation, State

(pooled)

Marriage/Cohabitation

Marital Status (CS)

Cohabitation Status (CS)

Marital Status Change (R)

Decennial Census All individuals All households and

individuals

•

Cross-sectional

Decennial

Available: 1790 – 2000;

1850 - 2000 (microdata)

Nation, State,

County, Census

Tract

Marital Status (CS)

Cohabitation Status (CS)

Early Childhood

Longitudinal Study -

Birth Cohort

(ECLS - B)

Children born in 2001 13,500 individuals

• Single panel

• Baseline: 2001 (Birth)

• Follow-up: 18-, 30-, and 48-

months old, kindergarten and first

grade

• Available: 2001

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD)

Cohabitation Status Change (PD)

Early Childhood

Longitudinal Study -

Kindergarten Cohort

(ECLS-K)

Children attending

kindergarten in fall 1998

22,000 individuals

• Single panel

• Baseline: 1998 (Kindergarten

Fall)

• Follow-ups: Kindergarten Spring,

, 3

, and 4

grades

• Available: 1998 -1999

Nation Marital Status (PD)

Cohabitati

on Status (PD)

Marital Status Change (PD)

Cohabitation Status Change (PD)

Fragile Families and

Child Wellbeing

Study

Mothers and fathers of

newly born children in

hospitals in cities with

populations over 200,000

people

4,700 families

(8,000 individuals)

• Single panel

• Baseline: 1998 - 2000 (Birth)

• Follow-ups: 1, 3, and 5 years

• Available: Baseline, 1 and 3 year

follow-up

Cities with

populations over

200,000 and 20

selected cities

Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD)

Cohabitation Status Change (PD)

Health and

Retirement Study

(HRS)

Individuals born before

1948 (non-

institutionalized at

baseline)

22,000 individuals

• Five single cohort panels

• Baselines: 1992, 1993, 1998,

2004

• Biennial follow-ups

• Available: 1992 - 2004

Nation Marital Status (PD)

Marital Status Change (PD, R)

Marital Instability

Over the Life

Course Study

Individuals, ages 18-55, in

intact marriages in 1980

2,000 individuals

• Single panel

• Baseline: 1980

• Follow-ups: 1983, 1988, 1992-

94, 1997, 2000

• All years available

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD, R)

Cohabitation Status Change (PD)

Medicare Current

Beneficiary Survey

(MCBS)

Individuals in Centers for

Medicare & Medicaid

Services (CMS)'s

Medicare enrollment file

12,000 individuals

• Four-year rotating panels (12

interviews)

• Annual data

• Available: 1991 - 2002

Nation Marital Status (PD)

Marital Status Change (PD)

Representative

Areas

Exhibit A.2: Summary of Other Data Sets

Name Population Sample Size

Survey Design

Representative

Areas

Marriage/Cohabitation

National Health

Interview Survey

(NHIS)

Non-institutionalized

civilian individuals

36,000 HH

•

Cross-sectional

Annual data

Available: 1957 - 2006

Nation Marital Status (CS)

Cohabitation Status (CS)

National

Longitudinal Survey

of Mature Women

(NLSMW)

Non-institutionalized

civilian women born

between 1923 and 1937

5,000 individuals

(currently 2,300

individuals)

•

Single panel

Baseline: 1967

Follow-ups: Every few years,

currently biennial

Available: 1967 – 2003

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD, R)

Cohabitation Status Change (PD)

National

Longitudinal Survey

of Young Women

(NLSYW)

Non-institutionalized

civilian women born

between 1942 and 1952

5,000 individuals

(currently 2,800

individuals)

•

Single panel

Baseline: 1968

Follow-ups: Every few years,

currently biennial

Available: 1968 – 2003

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD)

Cohabitation Status Change (PD)

National

Longitudinal Survey

of Youth - 1979

(NLSY79)

Non-institutionalized

individuals born between

1954 and 1964

12,600 individuals

(currently 8,000

individuals)

•

Single panel

Baseline: 1979

Follow-ups: Annual until 1994,

currently biennial

Available: 1979 – 2004

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD)

Cohabitation Status Change (PD)

National

Longitudinal Survey

of Youth - 1997

(NLSY97)

Non-institutionalized

individuals born between

1980 and 1984

9,000 individuals

•

Single panel

Baseline: 1997

Follow-ups: Annual

Available: 1997 – 2005

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD)

Cohabitation Status Change (PD)

National Survey of

America's Families

(NSAF)

Non-institutionalized

civilian individuals under

age 65

43,000 HH

(100,000

individuals)

•

Cross-sectional

Annual data

Available: 1997, 1999, 2002

Nation, 13 States Marital Status (CS)

Cohabitation Status (CS)

Marital Status Change (R)

National Survey of

Families and

Households (NSFH)

Non-institutionalized

individuals 19 and over

9,600 HH

(13,000

individuals)

•

Single panel

Baseline: 1987-88

Follow-ups: 1992-94, 2001-02

Available: 1987-88, 1992-94,

2001-02

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD, R)

Cohabitation Status Change (PD, R)

Panel Study of

Income Dynamics

(PSID)

Non-institutionalized

civilian individuals

8,000 families

(37,500

individuals)

•

Single panel

Baseline: 1968

Follow-ups: Annual until 1997,

currently biennial

Available: 1968-2005

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD, R)

Cohabitation Status Change (PD)

Survey of Program

Dynamics (SPD)

Non-institutionalized

civilian individuals

16,000 HH

•

Single panel

Baseline: 1992/93 SIPP panels

Follow-ups: Annual (1997 - 2002)

Available: 1992 - 2002

Nation Marital Status (PD)

Cohabitation Status (PD)

Marital Status Change (PD, R)

Cohabitation Status Change (PD)

Exhibit A.2: Summary of Other Data Sets

Representative

Name Population Sample Size Survey Design Marriage/Cohabitation

Areas

All information is current as of January 2005, except for selected updates where information is readily available.

The sample sizes presented in this table are those of the most recent data available as of January 2005, except for selected updates where information was readily available.

This table reports the representativeness of each survey at the national, state, and local levels. Several of these data sets are also representative for regions of the country.

PD = Panel Data, CS = Cross-Sectional Data, R = Retrospective Data.

The Current Population Survey included a supplement gathering retrospective marital histories roughly every 5 years from 1971 until 1995.

Marital status was only included in the 2000 Census long form, a survey of 1-in-6 households. Marital status was included on the short form prior to 2000, yielding block level estimates.

Researchers may use the Early Childhood Longitudinal Surveys to look at marriage/cohabitation of sample children's parents.

The Fragile Families and Child Wellbeing Study is representative of non-marital births in cities with over 200,000 people. There is debate over its representativeness of marital births.

This sample size is for all cohorts interviewed in 2002. The sample size of the cohort introduced in 1992 was 12,600 individuals. The sample of AHEAD, the cohort introduced in 1993, was 8,200.

The original sample size for all four cohorts in the 1998 survey was 22,000.

Two additional cohorts are scheduled to be introduced, one in 2004 and one in 2010.

While all the National Longitudinal Studies contain some retrospective data, the marital and cohabitation status change information is almost entirely from the longitudinal data.

Marital status change is only available for the 2002 NSAF.

VII. APPENDIX B: MARRIAGE AND DIVORCE QUESTIONS

This appendix presents the questions each of the three key data sets ask about marriage and

divorce. These questions are from the 2008 ACS, the 2002 NSFG, and the 2004 SIPP survey

questionnaires.

American Community Survey (ACS)

1. What is this person’s marital status?

1) Now married

2) Widowed

3) Divorced

4) Separated

5) Never married

2. In the past 12 months did this person get --

1) Married? (yes/no)

2) Widowed? (yes/no)

3) Divorced? (yes/no)

How many times has this person been married?

1) Once

2) Two times

3) Three or more times

4. In what year did this person last get married?

Year ______

National Survey of Family Growth (NSFG)

1. What is your current marital status?

1) Married

2) Not married but living together with a partner of the opposite sex

3) Widowed

4) Divorced

5) Separated, because you and your spouse are not getting along

6) Never be

en married

(Including your present marriage,) how many times have you been married?

Number ______

Questions from female questionnaire:

Following questions cover all marriages:

3. In what month and year were you an

d (HUSBAND) married?

4. How old were you when you got married (this [nth] time)?

Age in years ______

5. How old was (HUSBAND) when you got married?

Age in years ______

6. In what month and year was he born?

7. Some couples live together without being married. By living together, we mean having a sexual

relationship while sharing the same usual address. Did you and (HUSBAND) live together before you got

married?

1) Yes

2) No

8. In what month and year did you and he first start living together?

9. At the time you began living together, were you and he engaged to be married or have definite plans to

get married?

1) Yes

2) No

10. (Is/Was) (HUSBAND) Hispanic or Latino, or of Spanish origin?

(Asked only for the respondent’s 1st or current/separated husband)

1) Yes

2) No

11. Which of the groups on Card 2 describes (HUSBAND)'s racial background? Please select one or more

groups.

(Asked only for the respondent’s 1st or current/separated husband)

1) American Indian or Alaska Native

2) Asian

3) Native Hawaiian or

Other Pacific Islander

4) Black or African American

5) White

12. What is the highest level of education (HUSBAND) has completed?

(Asked only for current or separated husbands)

1) Less than high school

2) High school graduate or GED

3) Some college but no degree

4) 2-year college degree (e.g., Associate’s degree)

5) 4-year college graduate (e.g., BA,

BS)

6) Graduate

or professional school

13.

At the time you and he were married, had (HUSBAND) been married before?

1) Yes

2) No

14. When you and he got married, did he have any children, either biological or adopted, from any

previous relationships?

1) Yes

2) No

15.

How many children did he have?

Number _____

16. How did your (nth) marriage end?

(Asked if respondent is not married to or separated from this husband)

1) Death of husband

2) Divorce

3) Annulment

17. In what month and year did (HUSBAND) die?

18. In what

month and year did your (divorce become final/annulment take place)?

19. In what month and year did you and (HUSBAND) stop living together (for the last time)?

Questions from male questionnaire:

20.

Have you ever lived together with a female sexual partner? By living together, I mean having a sexual

relationship while sharing the same usual residence.

1) Yes

2) No

21.

How many female sexual partners have you lived with in your life?

Number ______

Following questions cover all marriages:

22. In what

month and year were you and she married?

23. How old were you when you and (WIFE) got married?

Age in years _______

24. Some couples live together without being married. By living together, we

mean having a sexual

relationship while sharing the same usual address. Did you and (WIFE) live together before you got

married?

1) Yes

2) No

25. In what month and year did you and she first start living together?

26. How old were you whe

n you and (WIFE) first started living together?

Age in years ________

27. At the time you first started living together with (WIFE), were you and she engaged to be married or

did you have definite plans to get married?

1) Yes

2) No

28. How did your marriage end?

1) Death of wife

2) Divorce

3) Annulment

4) Separation

29. In what month and year did (WIFE) die?

30. In what

month and year did your divorce become final?

31. In what month and year did your annulment take place?

32. In what month and year did you and (WIFE) last stop living together?

33. In what month and year was she born?

(Asked if this woman is the respondent’s current wife)

4. How old is (WIFE) now?

(Asked if this woman is the respondent’s current wife)

Age in years at last birthday ________

35. How old was (WIFE) when (she died/ your divorce became final/your annulment took place/ you and

she last stopped living together)?

Age in years ________

36. Was/Is (WIFE) Hispanic or Latino, or of Spanish origin?

(Asked if this woman was the respondent’s first wife)

1) Yes

2) No

37. Which of the groups describes (WIFE)’s racial background? Please

select one or more groups. ENTER

all that apply.

(Asked if this woman was the respondent’s first wife)

1) American Indian or Alaska Native

2) Asian

3) Native Hawaiian or Other Pacific Islander

4) Black or African American

5) White

38. What is the highest level of education (WIFE) has completed?

(Asked if this woman is the respondent’s current wife)

1) Less than high school

2) High school graduate or GED

3) Some college but no degree

4) 2-year college degree (e.g., Associate’s degree)

5) 4-year college graduate (e.g., BA,

BS)

6) Graduate

or professional school

39.

At the time you and she (were married/started living together), had she ever been married?

1) Yes

2) No

Survey of Income and Program Participation

1. What is your current marital status?

1) Married, spouse present

2) Married, spouse absent

3) Widowed

4) Divorced

5) Separated

6) Never married

2. How many times have you been married?

1) 1

2) 2

3) 3

4) 4+

Following questions cover up to 3 marriages:

3. In what month and year did you

get married for the (nth) time?

4. Did your (nth) marriage end in widowhood or divorce?

1) Widowhood

2) Divorce

5. In what month and year were you widowed?

6. In what month and year were you divorced?

7. In what month and year did you actually stop living with your (nth) spouse?