5
Often the goal of the study is not merely to report a
statistic—e.g., an estimate or confidence interval—but to
also draw a conclusion or make a decision based upon this
estimate. For example, in an explanatory question, the study
may conclude that an intervention is successful if there is
(a) evidence that the sample effect size estimate is > 0.20
standard deviations (or another threshold), and (b) that
there is evidence the effect is non-zero in the population
(e.g., a hypothesis test based on a Type I error of 0.05). In a
descriptive study, there may not be a decision per se, but
instead, the goal may be to precisely estimate a measure
(e.g., of poverty or achievement). In this case, a margin of
error might be specified. Finally, in a predictive study, a
new method or algorithm might be considered useful if it
has a lower mean-squared-error than an existing measure
or reaches a target accuracy metric. Importantly, these
criteria given here are only examples. Others might be more
appropriate depending on the case. What is important,
however, is that criteria are stated in advance. Table 2
below provides examples of these for each of the types of
questions.
Clear questions and goals shape research design and
analytic strategies. When misaligned, research projects fall
apart in their logic. For example, if a proposed explanatory
study seeks to determine if an intervention can improve
student learning broadly for all students, but then indicates
that the schools in the study are all found in a single large
urban school district, the results will not be able to meet the
proposed goal. Persuasive research proposals are careful,
therefore, to ensure that the questions asked are answerable
and stated inquiry goals are feasible using the proposed
research design and analysis methods.
3.0 Set Clear Goals
Offer a clearly defined and specific research question and
goals for the study, once the question type is identified.
This specification will ultimately narrow the question from
a broader class of interest for study—e.g., about inequality
in all schools—to the narrower class that can be feasibly
answered in the proposed study, given the budget and
real-world constraints. For example, while inequality may
be a broad concern, in a particular study one might focus
specifically on school funding inequality in North Carolina
elementary schools in 2018-2019. This specificity does not
make the research any less important—but it does help in
defining the scope of the claims that can be made from the
study.
In framing a study, it is important to define the population
to whom the results are intended to apply, and conversely,
where they may not. It is often helpful to turn to population-
level data, such as the Common Core of Data (an annual
census of schools), the American Community Survey, the U.S.
Census, or national probability survey data like the
Early Childhood Longitudinal Study (ECLS), to define the
characteristics of this population. For example, it may be that
the study focuses on rural students of color in the American
South, or that the study focuses on schools in Texas. Keep in
mind that the population definition is broader than simply
reporting summary statistics on the sample in hand (unless
the population of interest is the sample, which is possible).
Distinguishing between a sample and population often
involves asking questions like, “What is this sample a case
of?” and “Where might results of this study not apply?”
Once a target population is defined, offer a clear definition
of the parameter(s) of focus in the study. This requires
narrowing the question and anticipating the kinds of
‘output’ for focus in the quantitative analyses. For an
explanatory question, the goal may be to estimate the
average treatment effect, subgroup treatment effects, or
treatment effect differences. For a descriptive question,
the goal may be to provide summary statistics (e.g., means,
standard deviations, correlations), or the amount of variation
in an outcome explained by different subsets of variables.
For a predictive question, the goal may be to develop an
algorithm that can predict well (e.g., 95% accuracy) an
outcome for each student or school in the population. Again,
it is important to distinguish here between the parameter,
which is unknown in the population, and the estimator,
which is based on the model and data in the study.
A Guide to Quantitative Research Proposals