You are on page 1of 8

International Journal of Research & Method in Education

ISSN: 1743-727X (Print) 1743-7288 (Online) Journal homepage: http://www.tandfonline.com/loi/cwse20

Inferential statistics and the use of administrative


data in US educational research

Benjamin G. Gibbs, Kevin Shafer & Aaron Miles

To cite this article: Benjamin G. Gibbs, Kevin Shafer & Aaron Miles (2015): Inferential statistics
and the use of administrative data in US educational research, International Journal of
Research & Method in Education, DOI: 10.1080/1743727X.2015.1113249

To link to this article: http://dx.doi.org/10.1080/1743727X.2015.1113249

Published online: 29 Nov 2015.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=cwse20

Download by: [BYU Brigham Young University] Date: 30 November 2015, At: 10:30
INTERNATIONAL JOURNAL OF RESEARCH & METHOD IN EDUCATION, 2015
http://dx.doi.org/10.1080/1743727X.2015.1113249

Inferential statistics and the use of administrative data in US


educational research
Benjamin G. Gibbsa, Kevin Shaferb and Aaron Milesc
a
Department of Sociology, Brigham Young University, Provo, UT, USA; bSchool of Social Work, Brigham Young
University, Provo, UT, USA; cDepartment of Sociology, Ohio State University, Columbus, OH, USA
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

ABSTRACT ARTICLE HISTORY


While the use of inferential statistics is a nearly universal practice in the Received 19 August 2014
social sciences, there are instances where its application is unnecessary Accepted 2 September 2015
and potentially misleading. This is true for a portion of research using
KEYWORDS
administrative data in educational research in the United States. Statistics; inference;
Surveying all research articles using administrative data published in regression; administrative
Educational Researcher and American Educational Research Journal data
between 2011 and 2013, we find that over half use inferential
techniques to help interpret results despite the use of population data.
To the extent that these practices and findings are replicated and
interpreted in the United States and elsewhere, this can lead to the
undervaluing of potentially meaningful patterns and trends.

One strength of educational research is the availability of administrative data – largely due to federal
and state policies that often require school administrators to collect and track the progress of all stu-
dents in a given population (NCLB 2001; see Wayman, Cho, and Johnston 2007). Administrative data
may best be thought of as a census – data collected from all individuals in a defined population
whether at the national, state, district, school, or classroom level. In most circumstances the data
are not a sampling of students; yet US researchers often use p-values to indicate whether results
should be considered statistically (and practically) significant. We argue that this practice should
be used rarely, and when utilized, be clearly justified (e.g. superpopulations). Without proper justifi-
cation, the use of p-values could lead to nonreporting, errors in data interpretation, or undervaluing
of potentially important findings.

Unique data
In the United States, there are two policies that have aided in the collection and use of administrative
data in primary and secondary schools – the No Child Left Behind Act of 2001 (NCLB) and the Family
Educational Rights and Privacy Act of 1974 (FERPA). One purpose of NCLB was to ensure that schools
could be held accountable for the academic achievement of their students by using widespread
testing and proficiency measures (Bush 2001; NCLB 2001), which had the added benefit of facilitating
academic research (see Dworkin 2005; Kim and Sunderman 2005; Linn, Baker, and Betebenner 2002;
Wayman, Cho, and Johnston 2007). And although FERPA generally prevents schools from disclosing
student information without parental consent, data can be given out to ‘organizations conducting
studies for or on behalf of the school’ (U.S. Department of Education 1974).

CONTACT Benjamin G. Gibbs benjamin_gibbs@byu.edu


© 2015 Taylor & Francis
2 B. G. GIBBS ET AL.

Thus, in the United States and elsewhere, federal-level policies that incentivize the collection of
school-level population data provide an opportunity for educational researchers because various
data issues (e.g. non-random sampling and small sample sizes) can be avoided. Federal-, state-,
and school-level census data can be uniquely leveraged to understand important educational
issues and help develop educational policy as a result. Therefore, with access to data unconstrained
by sampling, often with a large number of cases, it is imperative that educational researchers properly
interpret and utilize this unique form of data.

Inferential statistics
Inferential statistics is most commonly a technique designed for sample – rather than population –
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

data. As one introductory statistics book states (Warner 2012):


Inferential statistics involve using a descriptive statistic for a sample to make inferences or estimates about the
value of a corresponding population parameter. For example, the use of a sample mean to test hypotheses
about the value of the corresponding population mean or to construct a confidence interval estimate for the
population mean. (1092)

Thus, inferential statistics are necessary tools to analyse sample data. Inference to the unknown
population parameter is possible due to the central limit theorem, which states that a given
random sample of a certain size can reasonably approximate an unknown population parameter.
Generally, inferential statistics is necessary because collecting population-level data is fraught with
difficulties from the technical (i.e. funding and time) to the conceptual (what constitutes a popu-
lation?). That this form of statistical analysis is well-known in education and other social sciences,
even covered sufficiently in introductory statistics textbooks (e.g. Warner 2012), does not prevent
its misuse.
Inferential statistics can account for estimation errors that are inherent in samples. Basic descrip-
tive statistics alone do not assign a level of confidence as to whether the sampled data match the
population (had it been known). The use of p-values and confidence intervals are needed to
provide some level of certainty that sample statistics infer to the larger population. Therefore,
results from sampled populations are estimates of the true population.1 A p-value of less than .05 indi-
cates that there is, at most, a 5% chance that the estimate does not accurately reflect the population
parameter due to sampling error. Or, in other words, that if 100 random samples of a given popu-
lation were taken, only 5 would produce a different estimate.
This debate is not new to education research. For example, Educational Researcher has published
lively debates over the use of p-values and the difference between statistical and practical signifi-
cance in the 1990s (see Robinson and Levin 1997; Thompson 1996, 1997). Thompson (1996)
argued that part of the issue results from a simple need for more statistical literacy:
Unfortunately, very few researchers seem to understand what their p calculated values actually evaluate … Put
succinctly, pcalculated is the probability (0 to 1.0) of the sample statistics, given the sample size, and assuming
the sample was derived from a population in which the null hypothesis is exactly true. (27)

In other words, a p-value is a statistic intended for sampled data and therefore use with census or
population data should be considered a departure from this basic assumption.
Are there instances where the use of inferential statistics for population data could be justified?
Could administrative data from Columbus city schools, for example, be used to estimate patterns
in all school districts in Ohio, the Midwest, or throughout the United States? Such extensions of
data are difficult to defend and are, in fact, rare in educational research. Columbus’ demographics
differ from Ohio’s in important ways. Thirty-two percent of Columbus city school residents are
African-American, compared to 12% of the state. The percentage of individuals in poverty is seven
percentage points higher in the city, and it is difficult to consider how a large urban school district
might somehow be representative of all school districts in a state containing numerous rural districts
INTERNATIONAL JOURNAL OF RESEARCH & METHOD IN EDUCATION 3

(U.S. Census Bureau 2013). Comparisons to other cities in Ohio would also be difficult – the demo-
graphics of Columbus, Cleveland, and Cincinnati are quite different from one another as would be
comparisons to past or future Columbus city schools.
Despite these problems, there are a few instances where administrative data from a particular
population has been justified to represent some larger population. A ‘superpopulation,’ for
example, can use sampled or census data to infer to a greater population that theoretically could
exist, may have existed, or may exist in the future. But given that there is debate regarding how
to use data for superpopulation inference (Graubard and Korn 2002; Western and Jackman 1994),
it is critical to identify this intention and defend its use (Ballinger 2011; Hartley and Sielken 1975;
Gibbs, Shafer, and Dufur 2015; Isaki and Fuller 1982). And although this technique could be used
to explain the use of inference with administrative data in educational research, this approach is
seldom cited or justified in the literature.
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

Evidence of misuse
To illustrate the scope of this issue, we searched articles published in Educational Researcher (ER) and
the American Educational Research Journal (AERJ) (two top educational research journals in the United
States) between 2011 and 2013. These journals represent highly visible generalist journals and thus
have a potential impact in shaping the field. Both are sponsored by American Education Research
Association (AERA). AERA is a large education association with an annual attendance at conferences
of over 14,400 attenders and 10,000 presenters (in 2014).2 Also, ER is ranked 7 out of 219 educational
journals and has a 2013 impact factor of 2.963. AERJ is ranked 15 out of 219 and has a 2013 impact

Table 1. Summary of articles in ER and AERJ that use census administrative data 2011–2013.
Interpret
Authors Year Journal Population p-values?
Henry, Bastian, and 2011 Educational Researcher Early North Carolina Teachers Yes
Fortner
Pang, Han, and Pang 2011 Educational Researcher K-12 Students in California 2003–2008 Yes
Daly and Finnigan 2011 American Educational Research Administrators in ‘La Estasis School Yes
Journal District’
Papay 2011 American Educational Research Teachers in Large Urban School District No
Journal
Carey and Carifio 2012 Educational Researcher Students in ‘Mill City High School’ Yes
2003–2010
Fantuzzo et al. 2012 Educational Researcher Elementary Students in Philadelphia Yes
District
Henry, Bastian, and Smith 2012 Educational Researcher North Carolina Teachers Yes
McEachin and Polikoff 2012 Educational Researcher California Schools 2005–2011 Yes
Robinson and Espelage 2012 Educational Researcher Schools in Dane County, WI Yes
Voight, Shinn, and Nation 2012 Educational Researcher Students in a Tennessee school district Yes
Herbers et al. 2012 Educational Researcher Students in Minneapolis Public Schools No
Maruyama 2012 Educational Researcher Students in Minnesota Public Schools No
Warren and Saliba 2012 Educational Researcher American Students No
Ronfeldt, Loeb, and 2012 American Educational Research 4th and 5th Graders in New York City Yes
Wyckoff Journal Schools
Boyd et al. 2012 American Educational Research Teachers in New York City Schools Yes
Journal
Long, Conger, and 2012 American Educational Research Florida students and teachers Yes
Iatarola Journal
Nomi and Allensworth 2012 American Educational Research Chicago School District 9th graders Yes
Journal
VanDerHeyden et al. 2012 American Educational Research 4th and 5th graders in specific school Yes
Journal district
Flores and Park 2013 Educational Researcher Students in Texas Schools Yes
Kalogrides and Loeb 2013 Educational Researcher Students in three school districts Yes
Winters and Cowen 2013 Educational Researcher Florida students and teachers No
4 B. G. GIBBS ET AL.

factor of 2.275. Other highly visible journals in education are often more specialized (e.g. math edu-
cation and elementary school education).3 To the extent that these journals are influential sources for
educational scholars, we feel our coverage of these two journals, even though based in the United
States, warrants concern for both US and international scholars.
Table 1 documents all cases we coded as using census administrative data in ER and AERJ from
2011 to 2013 and we indicate whether these studies interpreted p-values in their analyses. In
2011, four articles used population data of which three potentially misused inferential statistics. In
2012, of the 14 articles that use population data, we document 11, which used inferential statistics.
And in 2013, of the three articles that used population data, two potentially misused inferential stat-
istics. Overall, of the 21 articles with census population data, 16 used inferential statistics to interpret
results.
We should also note that the number of articles that use administrative data fluctuates by year. Of
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

the 54 research articles published in 2011 in both EA and AERJ, 4 used administrative data. In 2012, of
the 55 articles published, 14 used administrative data and in 2013, 63 articles were published, with 3
using administrative data.
Henry, Bastian, and Smith’s (2012) study is a good example of how analyses can potentially be
misused. The study utilizes census data collected from North Carolina public schools between the
years of 2005–2006 and 2009–2010 for new teachers (in their first five years) to examine the effect
of merit-based scholarships on a variety of early teaching outcomes. The authors used t-tests,
multi-level modeling, and logistic regression. The authors interpreted p-values as evidence of statisti-
cally significant differences in their models. Yet, the justification for the use of inferential statistics was
not presented. As the authors used population data, it remains unclear which larger population these
data are intended to infer – early career teachers in other states? Later career teachers? Early career
teachers in the past or future of North Carolina public schools? Furthermore, even if the data were to
generalize to other school districts, there remains one critical problem – using a non-random ‘sample’
to infer to a larger population.
Without a clearly defined population (i.e. all teachers in the United States or all teachers in North
Carolina), these analytic methods are without sufficient justification. While results of Henry, Bastian,
and Smith’s (2012) study may be of interest to scholars and warrant further investigation in other
states and contexts, under most circumstances these findings should not be inferred to any larger
population because North Carolina is not likely a sample of any other population. This problem is
not simply an unfortunate anomaly by the authors in a top education journal – previous studies con-
ducted by some of the same authors repeat the ambiguity (see Henry, Bastian, and Fortner 2011). In
sum, as we document the widespread use of inferential statistics with census data in these two
important journals, the assumptions embedded in its use are often unacknowledged and
unspecified.

The consequences
Education research is unique in that many articles have direct policy implications, especially those
that use administrative and census data. Because one requisite for access to administrative data
under FERPA is that research is conducted for the benefit of the school (U.S. Department of Education
1974), administrators and educators are likely to seek ways to implement policies from research find-
ings. If findings are ignored due to low confidence estimates (for which there is no population to esti-
mate), this may have a direct impact on students, teachers, and administrators. Thus, there may be
findings of value in the educational research community yet to be reported due to lack of supposed
statistical significance.
To illustrate, Henry, Bastian, and Fortner (2011) draw conclusions from their interpretation of
p-values in potentially misleading ways. In at least five occasions in the results section, the authors
downplay or disregard findings about teacher effectiveness by concluding that ‘results are not statisti-
cally significant.’ Thus, results that show evidence that teachers vary in their effectiveness in the
INTERNATIONAL JOURNAL OF RESEARCH & METHOD IN EDUCATION 5

classroom are relegated to non-significance. In total, the authors claim that there are no significant
differences in the following: teacher effectiveness from third and fifth years of teaching, between
elementary school teachers who leave after their first year and those who stay for a longer period of
time, the reading effectiveness for second year middle school teachers over time, the decline in
fourth year effectiveness for middle school teachers who leave after four years, and differences in
teacher effectiveness between the third and fifth years of teaching for high school teachers overall.
Taken together, the overreliance on p-values leads these authors to dismiss several potentially impor-
tant findings because they did not cross the threshold of statistically significant. In conventional cases,
we argue that it is important to emphasize that census data can reveal patterns of the data for which
statistical inference is not needed to aid in its interpretation. Of course, effect sizes are important to
consider and there are still unresolved questions regarding measurement validity and reliability, but
p-values and related inferential techniques are simply not the appropriate tool to address these issues.
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

Like Henry, Bastian, and Fortner (2011), other articles we analysed reveal potential misuse of infer-
ential statistics with census data. For example, Fantuzzo et al. (2012) dismiss the difference in math and
reading achievement between students who are homeless (but stay in the same school) and students
who are homeless and frequently move between schools. Carey and Carifio (2012) discard the effect of
minimum grading policies on passing grades when examined across several years. The authors state
that ‘it remains to be seen whether the patterns at this high school exist at a national level,’ but interpret
p-values as if their results generalize to this or another unspecified population. And Flores and Park’s
(2013) conclude that the effects of race on college completion disappear (statistically) when account-
ing for institutional characteristics even though the coefficients still indicate a relationship.4
To illustrate how these examples could negatively shape future research, consider how a program
impact analysis between two different schools, one with 200 students, and the other with 2000 stu-
dents, might be incorrectly executed. As t-tests, p-values, and confidence intervals rely on the
assumption that the data are a sample,5 the researcher who uncritically uses these techniques
may come to the conclusion that the program is effective only for large schools, even when the
program may have had the same impact on small schools. Of course, census data in a smaller
school may have more variation (e.g. means are more sensitive to outliers), but this and related
issues are problems of understanding the data, not ones that inferential statistical techniques are tra-
ditionally developed to address.

Conclusion
Why is the use of inferential statistics so prevalent in education research? One explanation may be the
uncritical application of statistical techniques, as has been the case in various disciplines (i.e. Ballinger
2011; Gibbs, Shafer, and Dufur 2015; Leonard and Phillips 1997; Sampson and Laub 2005) and in edu-
cational research (Robinson and Levin 1997; Thompson 1996, 1997). Another is that education
researchers regularly use various inferential analyses when examining large sample datasets such
as the National Longitudinal Survey of Adolescent Health (Add Health), The National Educational
Longitudinal Study (NELS), The Early Childhood Longitudinal Study (ECLS), and others. As many influ-
ential studies in education research properly apply inferential statistics on large datasets such as
these, researchers may uncritically apply these same methods when using administrative and
census data. As we show, this can lead researchers to devalue potentially important findings and
overvalue others. Whether this is the result of ambiguity or error, census data in educational research
requires proper use to realize the unique potential of this form of data.6
We suggest educational research journals adopt or revise standards to make the interpretation of
inferential statistics and p-values, such as defining the intended inferred population, a required part
of reporting in analytic methods sections (American Educational Research Association 2006). With a
movement towards quantitative literacy and proficiency in the social sciences (Henson, Hull, and Wil-
liams 2010), analyses should conform to the basic assumptions of inferential statistics, especially
when using census data.
6 B. G. GIBBS ET AL.

Notes
1. Although there may be other forms of error common to both sample and population data, such as measurement
validity and reliability, the issue of inference is solely concerned about confidence in generalizing findings.
2. See http://www.aera.net/EventsMeetings/AnnualMeeting/tabid/10208/Default.aspx.
3. Of international journals, we found no journals that were both generalist and highly visible journals (e.g. International
Journal of Sustainability in Higher Education, International Journal for Educational and Vocational Guidance are highly
ranked but specific in scope). Also, when examining international educational journals, we found few instances of
regular misuse of inferential statistics like those found in ER and AERJ.
4. The interpretation of p-values is not always detrimental. In many cases, results are relatively unchanged (see Boyd
et al. 2012), but the use of inferential statistics still needs explicit justification.
5. The t-value is estimated by taking the difference of the observed coefficient b and the expected coefficient b and
dividing that by the standard error. The standard error is a function of sample size, with the assumption that as
the sample size increases, the sample mean will more accurately reflect the population mean. Therefore, as the
sample size increases (assuming the standard deviation remains constant), results from t-tests increase, p-values
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

decrease, and confidence intervals tighten.


6. As a reviewer points out, we should note that this does not suggest a movement towards simple analyses of census
data. There are inherent problems even with administrative data that require sophisticated statistical techniques.

Disclosure statement
No potential conflict of interest was reported by the authors.

References
American Educational Research Association. 2006. “Standards for Reporting on Empirical Social Science Research in AERA
Publications.” Educational Researcher 35 (6): 33–40.
Ballinger, C. 2011. “Why Inferential Statistics are Inappropriate for Development Studies and How the Same Data can be
Better Used.” http://ideas.repec.org/p/pra/mprapa/29780.html.
Boyd, D., P. Grossman, K. Hammerness, H. Lankford, S. Loeb, M. Ronfeldt, and J. Wyckoff. 2012. “Recruiting Effective Math
Teachers: Evidence from New York City.” American Educational Research Journal 49 (6): 1008–1047.
Bush, G. 2001. “No Child Left Behind Foreward by President.” United States. http://www.ed.gov/nclb/landing.jhtml.
Carey, T., and J. Carifio. 2012. “The Minimum Grading Controversy: Results of a Quantitative Study of Seven Years of
Grading Data from an Urban High School.” Educational Researcher 41 (6): 201–208.
Daly, A. J., and K. S. Finnigan 2011. “The Ebb and Flow of Social Network Ties Between District Leaders Under High-Stakes
Accountability.” American Educational Research Journal, 48 (1): 39–79.
Dworkin, A. G. 2005. “The No Child Left Behind Act: Accountability, High-Stakes Testing, and Roles for Sociologists.”
Sociology of Education 78 (2): 170–174.
Fantuzzo, J. W., W. A. LeBoeuf, C.-C. Chen, H. L. Rouse, and D. P. Culhane. 2012. “The Unique and Combined Effects of
Homelessness and School Mobility on the Educational Outcomes of Young Children.” Educational Researcher 41 (9):
393–402.
Flores, S. M., and T. J. Park. 2013. “Race, Ethnicity, and College Success: Examining the Continued Significance of the
Minority-Serving Institution.” Educational Researcher 42 (3): 115–128.
Gibbs, B. G., K. Shafer, and M. J. Dufur. 2015. “Why Infer? The use and Misuse of Population Data in Sport Research.”
International Review for the Sociology of Sport 50 (1): 115–121.
Graubard, B. I., and E. L. Korn. 2002. “Inference for Superpopulation Parameters using Sample Surveys.” Statsitical Science
17 (1): 73–96.
Hartley, A. H. O., and R. L. Sielken Jr. 1975. “A “Super-Population Viewpoint” for Finite Population Sampling.” Biometrics 31
(2): 411–422.
Henry, G. T., K. C. Bastian, and C. K. Fortner. 2011. “Stayers and Leavers: Early-Career Teacher Effectiveness and Attrition.”
Educational Researcher 40 (6): 271–280.
Henry, G. T., K. C. Bastian, and A. A. Smith. 2012. “Scholarships to Recruit the “Best and Brightest” into Teaching: Who is
Recruited, Where Do They Teach, How Effective are They, and How Long Do They Stay?” Educational Researcher 41 (3):
83–92.
Henson, R. K., D. M. Hull, and C. S. Williams. 2010. “Methodology in our Education Research Culture: Toward a Stronger
Collective Quantitative Proficiency.” Educational Researcher 39 (3): 229–240.
Herbers, J. E., J. J. Cutuli, L. M. Supkoff, D. Heistad, C. Chan, E. Hinz, and A. S. Masten. 2012. “Early Reading Skills and
Academic Achievement Trajectories of Students Facing Poverty, Homelessness, and High Residential Mobility.”
Educational Researcher 41 (9): 366–374.
INTERNATIONAL JOURNAL OF RESEARCH & METHOD IN EDUCATION 7

Isaki, C. T., and W. A. Fuller. 1982. “Survey Design Under the Regression Superpopulation Model.” Journal of the American
Statistical Association 77 (377): 89–96.
Kalogrides, D., and S. Loeb. 2013. “Different Teachers, Different Peers: The Magnitude of Student Sorting within Schools.”
Educational Researcher 42 (6): 304–316.
Kim, J. S., and G. L. Sunderman. 2005. “Measuring Academic Proficiency Under the No Child Left Behind Act: Implications
for Educational Equity.” Educational Researcher 34 (8): 3–13.
Long, M. C., D. Conger, and P. Iatarola. 2012. “Effects of High School Course-Taking on Secondary and Postsecondary
Success.” American Educational Research Journal 49 (2): 285–322.
Leonard, W. M., and J. Phillips. 1997. “The Cause and Effect Rule for Percentaging Tables: An Overdue Statistical Correction
for Stacking Studies.” Sociology of Sport Journal 14 (3): 283–289.
Linn, R. L., E. L. Baker, and D. W. Betebenner. 2002. “Accountability Systems: Implications of Requirements of the No Child
Left Behind Act of 2001.” Educational Researcher 31 (9): 3–16.
Maruyama, G. 2012. “Assessing College Readiness: Should we be Satisfied with ACT or other Threshold Scores?”
Educational Researcher 41 (7): 252–261.
McEachin, A., and M. S. Polikoff. 2012. “We are the 5%: Which Schools would be Held Accountable under a Proposed
Downloaded by [BYU Brigham Young University] at 10:30 30 November 2015

Revision of the Elementary and Secondary Education Act?” Educational Researcher 41 (7): 243–251.
No Child Left Behind Act of 2001, Pub. L. No. 107–110, 115 Stat. 1425.
Nomi, T., and E. M. Allensworth. 2012. “Sorting and Supporting: Why Double-Dose Algebra Led to better Test Scores but
more Course Failures.” American Educational Research Journal 50 (4): 756–788.
Pang, V. O., P. P. Han, and J. M. Pang. 2011. “Asian American and Pacific Islander Students: Equity and the Achievement
Gap.” Educational Researcher 40 (8): 378–389.
Papay, J. P. 2011. “Different Tests, Different Answers: The Stability of Teacher Value-Added Estimates Across Outcome
Measures.” American Educational Research Journal 48 (1): 163–193.
Robinson, D. H., and J. R. Levin. 1997. “Research News and Comment: Reflections on Statistical and Substantive
Significance, With a Slice of Replication.” Educational Researcher 26 (5): 21–26.
Robinson, J. P., and D. L. Espelage. 2012. “Bullying Explains Only Part of LGBTQ–Heterosexual Risk Disparities Implications
for Policy and Practice.” Educational Researcher 41 (8): 309–319.
Ronfeldt, M., S. Loeb, and J. Wyckoff. 2012. “How Teacher Turnover Harms Student Achievement.” American Educational
Research Journal 50 (1): 4–36.
Sampson, R. J., and J. H. Laub. 2005. “Seductions of Method: Rejoinder to Nagin and Tremblay’s ‘Developmental Trajectory
Groups: Fact or Fiction?’” Criminology 43 (4): 905–913.
Thompson, B. 1996. “AERA Editorial Policies Regarding Statistical Significance Testing: Three Suggested Reforms.”
Educational Researcher 25 (2): 26–30.
Thompson, B. 1997. “Rejoinder: Editorial Policies Regarding Statistical Significance Tests: Further Comments.” Educational
Researcher 26 (6): 29–32.
U.S. Census Bureau. 2013. Statistical Abstract of the United States, 2012. Washington, D.C: U.S. Department of Commerce.
U.S. Department of Education. “Family Education Right and Privacy Act 1974.” http://www2.ed.gov/policy/gen/guid/fpco/
ferpa/index.html.
VanDerHeyden, A., T. McLaughlin, J. Algina, and P. Snyder. 2012. “Randomized Evaluation of a Supplemental Grade-Wide
Mathematics Intervention.” American Educational Research Journal 49 (6): 1251–1284.
Voight, A., M. Shinn, and M. Nation. 2012. “The Longitudinal Effects of Residential Mobility on the Academic Achievement
of Urban Elementary and Middle School Students.” Educational Researcher 41 (9): 385–392.
Warner, R. W. 2012. Applied Statistics: From Bivariate through Multivariate Techniques. (2nd ed.). Los Angeles, CA: Sage.
Warren, J. R., and J. Saliba. 2012. “First-Through Eighth-Grade Retention Rates for All 50 States A New Method and Initial
Results.” Educational Researcher 41(8): 320–329.
Wayman, J. C., V. Cho, and M. T. Johnston. 2007. The Data-Informed District: A District-Wide Evaluation of Data Use in the
Natrona County School District. Austin: The University of Texas.
Western, B., and S. Jackman. 1994. “Bayesian Inference for Comparative Research.” The American Political Science Review
88 (2): 412–423.
Winters, M. A., and J. M. Cowen. 2013. “Who Would Stay, Who Would be Dismissed? An Empirical Consideration of Value-
Added Teacher Retention Policies.” Educational Researcher 42 (6): 330–337.

You might also like