Understanding Student Self-Reports of Academic Performance and

AERA Open

April-June 2017, Vol. 3, No. 2, pp. 1 –14

DOI: 10.1177/2332858417711427

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons

Attribution-NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial

use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and

Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage).

Surveys are one of the most common ways to collect infor-

mation about students. They are used extensively in aca-

demic educational research, primarily to collect information

about students that is not easily accessible. Student surveys

also play an increasingly large role in accountability efforts

in many school districts. Perhaps the best known of these

student surveys is the Tripod 7C

(Ferguson, 2008, 2012), in

which students respond to questions about their teacher and

their classroom environment. The Tripod 7C has been

administered in school districts for over a decade, resulting

in, according to the developers, data being collected from

millions of students in every region of the United States.

More recently, the CORE Districts in California are imple-

menting a new School Quality Ratings Index

that heavily

weighs self-reported student data, while the New York City

school district administers the NYC School Survey

to stu-

dents in Grades 6–12 to understand more about the learning

environment.

Academic researchers and school administrators conduct

surveys of students to collect information not available in

administrative or other databases. Yet our heavy reliance on

student survey data raises questions about their accuracy.

While many student survey questions are attitudinal (e.g., “I

get nervous in this class”), academic and administrative sur-

veys ask students factual questions about themselves, their

classrooms, and their schools. Factual questions are distin-

guished from attitudinal questions in that they have a clear

and correct answer. For example, responses to questions

about a student’s overall grade point average (GPA) or

whether his or her teacher asks questions during class to be

sure that students are following along while he or she teach-

ing could be compared with data from administrative data-

bases or classroom observations. While attitudinal questions

query the respondents about their states of mind and thus

cannot be verified, responses to factual questions have cor-

rect answers and can be independently verified.

This raises an important question: To what extent do

K–12 students accurately respond to these types of ques-

tions? Despite a growing reliance on students reporting fac-

tual information about themselves and their classrooms in

school practice and academic research, we know quite little

about the accuracy of student self-reports, particularly

related to the specific types of questions on which students

tend to inaccurately report. The literature in this area is

largely confined to comparing self-reports of GPA and stan-

dardized test scores with administrative databases. Yet even

a cursory review of surveys such as the Tripod 7C and sur-

veys of secondary students conducted by the National Center

for Education Statistics that are often used by academic

researchers reveals that we ask students many more types of

Understanding Student Self-Reports of Academic

Performance and Course-Taking Behavior

Jeffrey A. Rosen

RTI International–Chicago

Stephen R. Porter

North Carolina State University

Jim Rogers

RTI International–Research Triangle Park

In recent years, student surveys have played an increasingly large role in educational research, policy making, and, particu-

larly, accountability efforts. However, research on the accuracy of students’ self-reports about themselves and their education

is limited to analyses of overall grade point average and ACT/SAT standardized test scores. Using a unique data set, we

investigate the accuracy of students’ survey responses to questions about their course taking and grades in mathematics dur-

ing high school. We then analyze which student and survey characteristics influence accuracy. We find that students are

reasonably good reporters of course-taking patterns but poor reporters of more potentially sensitive questions, including

when the student completed Algebra I and the grade earned in the course. We find that lack of accuracy in student survey

reports is consistently related to several student characteristics.

Keywords: educational policy, measurements, policy, self reports, survey research, validity/reliability

711427EROXXX10.1177/2332858417711427Rosen et al.Accuracy of Student Self-Reports

research-article2017

Rosen et al.

factual questions: questions about themselves, actions that

they have taken in schools, and when they took these actions.

The purpose of this study is to use a unique data source to

understand more about the ability of students to report fac-

tual information when surveyed. We used the High School

Longitudinal Study (HSLS:09) to compare student survey

responses to questions about their academic records with

information from their school course transcript data. To our

knowledge, this data source is one of the few available in

which students’ survey responses can be compared with

administrative databases (which are presumed to be accu-

rate). Prior studies have focused largely on GPA and test

score reporting accuracy. This study adds course-taking and

specific course grade reporting accuracy to the field’s knowl-

edge of misreporting in student surveys. We also investigate

the specific student characteristics that explain reporting

accuracy. Our analyses shed light on the extent to which stu-

dents can accurately report critical academic information

beyond the commonly studied GPA and SAT/ACT score.

Prior Research

Accuracy of Self-Reports

It is common practice in education research to use self-

reports of student grades in research studies. Since a widely

cited meta-analysis by Kuncel, Credé, and Thomas (2005),

many researchers have argued that self-reported grades are

generally accurate (i.e., Ratelle & Duchesne, 2014) and can

be safely used as measures of student performance in studies

of educational outcomes and interventions. Today, the prac-

tice of using self-reports of grades and scores remains typical

even among very well-known researchers writing in the top

educational research journals (i.e., Guo, Marsh, Morin,

Parker, & Kaur, 2015; Yeager et al., 2016). However, a review

of the research over the last 15 years suggests that student

self-reports, particularly of factual questions such as GPA

and test scores, may suffer from systematic inaccuracy.

Studies of self-reported student data have a fairly lengthy

history. Cautionary notes on students’ ability to report grades

and test scores accurately were raised early on by Maxwell

and Lopus (1994): They found students with below-average

grades to be most likely to misreport, and this finding has

been frequently replicated over the years. Cassady (2001), in

a sample of undergraduate students, found the lowest-per-

forming students (lowest quartile) to be much less accurate

reporters than students in higher-performance categories.

Zimmerman, Caldwell, and Bernat (2002) found widespread

self-reporting inaccuracy and a tendency by lower-perform-

ing students to overreport GPA by at least two half grades.

Mayer et al. (2007) found systematic overreporting in SAT

scores, particularly among lower-scoring students. Cole and

Gonyea (2010) investigated reporting accuracy in self-

reported ACT scores, finding that when students are inac-

curate in reporting their scores, a disproportionate number of

them overreport their scores; again, lower-achieving stu-

dents are much less accurate when reporting their scores.

Despite the issue with misreporting among lower-per-

forming students, early studies on the use of self-reported

measures of academic performance offered some hope in

that they seemed to demonstrate that self-reports show

acceptable accuracy. Cassady (2001) found self-reported

GPA in a sample of college students to be accurate, based on

correlations between self-reported GPA and official univer-

sity records (r = .97). The highly influential meta-analysis

by Kuncel et al. (2005) revealed that self-reported perfor-

mance measurements such as GPA and SAT score are gener-

ally accurate and can be safely used when administrative

data are unavailable.

Recently, researchers have taken a new look at self-

reporting accuracy. Caskie, Sutton, and Eckhardt (2014)

investigated reporting accuracy in a sample of undergradu-

ates, finding that females on average overreported their

actual college GPA and males underreported it but only in

the lowest-performing groups. Teye and Peaslee (2015)

studied grades and attendance reporting among younger stu-

dents, finding student reports to be inaccurate, especially

among lower-performing children. Schwartz and Beaver

(2015) offer some very compelling recent data on self-

reporting accuracy, using the National Longitudinal Study of

Adolescent Health. The authors found that self-reported

GPA was approximately one-half letter grade greater than

GPA recorded in administrative databases, and again, inac-

curacy was greater for lower-performing students.

One reason why many scholars erroneously conclude that

student self-reports are accurate is due to an overreliance on

bivariate correlations. Take, for example, the Kuncel et al.

(2005) finding that the correlation between actual and self-

reported GPA is .82. This relationship sounds robust, until

one estimates the percentage of variance in self-reports due

to actual GPA. If we regressed students’ self-reported GPA

on actual GPA, an r of .82 indicates that the R

from this

regression model would be .67. In other words, actual GPA

explains only two-thirds of the variance in self-reported

GPA. From this perspective, self-reports appear to contain

substantial error. Of course, if this were random error, we

would be less concerned, but the literature is clear that this is

not the case: There are substantial patterns of overreporting

among lower-performing students.

The accuracy of student reports on course taking has been

studied much less extensively than grades and test scores. In

general, authors seem to have disseminated findings through

technical reports. The ACT registration section, which

includes items for students to report grades and courses

taken, has been used to assess the accuracy of course-taking

self-reports. It seems that students taking the ACT report

their courses highly accurately: Valiga (1986) found accu-

rate reporting for course taking to be 95%, and Sawyer,

Laing, and Houston (1988) found it to be 87%. As recently

Accuracy of Student Self-Reports

as 2015 (Sanchez & Buddin, 2015), the ACT data sources

have been used, again revealing very high rates (>90%) of

course reporting accuracy.

Peer-reviewed articles on the accuracy of self-reporting

courses taken seem to be quite rare. Niemi and Smith (2003)

provide one notable analysis of course-taking accuracy,

using data from the 1994 National Assessment of Educational

Progress and the 1994 High School Transcript Study. They

found that students dramatically overstated the number of

history classes taken and failed to distinguish among differ-

ent types of history classes. More peer-reviewed work on

self-reports of course taking is needed.

Why Self-Reports May Be Inaccurate

Why might students be inaccurate reporters of their aca-

demic records? The survey methodology literature offers

some insights. Tourangeau, Rips, and Rasinski (2000) devel-

oped a model of the survey response process consisting of

four major components: comprehension (understanding

what is being asked), retrieval (being able to retrieve from

memory information import to forming a response), judg-

ment (aggregating all retrieved information and coming up

with an answer), and response (the actual reporting of the

answer). Error is possible at any of these points during the

survey response process.

Retrieval is one area where the response process for stu-

dents could break down. Retrieval success for academic

events will depend on their distinctiveness, when they

occurred, and whether respondents are asked to report on

events that occurred within specific time boundaries. More

distinctive events are more likely to be encoded in memory,

which results in their ability to be recalled. One issue with

surveys about academic behavior is that much of what we

might ask of students is not distinctive, unless it is out of the

ordinary. One example is course grades: With so many

courses throughout the academic career, students will face

difficulties in reporting course grades unless, for example,

they usually receive A’s but receive a D or vice versa.

Retrieval also depends on the effort put into the retrieval

process. Thus, students who have difficulty focusing or are

uninterested in the survey topic will likely, on average,

devote less time and effort into retrieving the requested

information, with subsequent higher error rates.

At the judgment stage, students must take all of the infor-

mation that they have retrieved from memory and construct

an answer. Often memories are not complete, and students

will infer an answer from partial memories. If students can-

not recall taking a course, they may interpret the lack of

memory as having not taken the course and so report a “no”

rather than “don’t know” response. Because it is difficult to

recall the timing of a particular event, students may guess

when they have taken part in an academic activity.

Once respondents have an answer to report, they face two

additional decisions: First, they must determine how to map

the response onto the response scale provided by the survey;

second, they must decide whether to alter the response. The

first issue tends to occur with questions that use vague

response scales, such as often/very often or agree/strongly

agree, where the meanings of the response categories are

unclear. The second can occur with any survey question that

asks about potentially embarrassing information, with some

respondents altering the response to give the socially desir-

able answer (Duckworth & Yeager, 2015.)

Two other causes of misreporting and inaccuracy could

be mischievous or careless behavior among survey respon-

dents, which may be particularly problematic among adoles-

cents. Robinson-Cimpian (2014) defines the mischievous

respondent as one who enters responses that she or he thinks

are funny (i.e., reporting they are adopted when they are not)

or implausibly extreme on items related to, for example,

alcohol consumption. One serious issue with the mischie-

vous respondent is the potential impact on subgroup esti-

mates. Robinson-Cimpian shows how a relatively small

number of mischievous respondents can introduce bias in

subgroup estimates of characteristics such as disabilities and

gender identity. By removing the mischievous respondents,

he shows how estimates of some characteristics can be

changed, suggesting that mischievous responses introduce

systematic bias.

Carelessness (random or thoughtless answers) on the sur-

vey task can also introduce error, leading to inaccurate sur-

vey responses. One common method to identify careless

respondents involves introducing survey items specifically

designed to uncover carelessness. For example, a survey

may introduce nonsense items or may place a series of

effort-based questions at the end of the substantive content

sections of the survey (Meade & Craig, 2012). These

approaches have the downside of lengthening survey admin-

istration, so it may be more advisable to correct for careless-

ness or a lack of survey effort post hoc. An interesting post

hoc approach to identifying low-effort survey respondents

involves examining item nonresponse. Since it is long estab-

lished that survey nonresponse is not random (Krosnick &

Presser, 2010) and often reflects underlying attributes of sur-

vey respondents, Hitt, Trivitt, and Cheng (2016) argue that

item missingness can be used as a proxy measure for effort

on the survey task. In a study based on 6 nationally represen-

tative data sets, Hitt et al. used item missingness as a proxy

for effort and conscientiousness, finding the percentage of

items skipped on a survey to be a significant predictor of

educational outcomes later in life. The clear implication

from Hitt et al. is that effort on the survey task, as reflected

in item missingness, does reflect something important about

the survey respondent.

It seems fairly clear that self-reported student GPA and

test scores often appear to suffer from systematic inaccuracy,

yet use of self-reported grades in educational research is a

fairly standard practice. Recent studies of certain socioemo-

tional outcomes (i.e., Feldman & Kubota, 2015; Guo et al.,

Rosen et al.

2015; Yeager et al., 2016), gender-based motivation in math

and science (Diseth, Meland, & Breidablik, 2014; Leaper,

Farkas, & Brown, 2012), the effects of working on academic

performance (Darolia, 2014), and adjustment in school

(Ratelle & Duchesne, 2014) all use self-reported measures

of academic performance. The widespread use of self-

reported grades and scores is most likely due to the ease and

relative inexpensiveness of collecting these data, especially

when compared with the challenges of collecting adminis-

trative records data. However, it seems necessary to not only

renew old cautions about using self-reported measures but

better understand why self-reporting inaccuracy on these

types of questions seems so common for students.

Furthermore, it seems wise to examine different types of

questions (i.e., specific course grades and courses taken)

commonly posed to students to determine if they, too, suffer

from systematic inaccuracy.

Research Questions

Possible sources of error may be present in student self-

reporting. While the literature has suggested a potential lack

of accuracy in self-reports of GPA and test scores for some

time, it has not offered guidance about the data quality of

other types of factual questions. In addition to adding new

analyses of the accuracy of self-reporting grades based on a

unique data set, we seek to contribute to the literature by

adding course grades and course taking to the evidence pool

on student self-reporting.

Specifically, our paper seeks to answer the following

research questions:

Research Question 1: How accurate are student self-

reports of courses taken and academic performance?

Research Question 2: Are there systematic patterns in the

direction of error? In other words, do students tend to

overreport positive outcomes and underreport nega-

tive outcomes?

Research Question 3: How do student characteristics and

aspects of the survey explain self-report accuracy?

The questions that we address here are critical for research

and practice for two reasons. First, one alternative to student

self-report data collection is collecting factual information

from student transcripts or other administrative records.

However, transcript collection tends to be cost prohibitive

for most researchers, leaving them little recourse but to rely

on self-reported information. For these users of data, more

information on self-reporting accuracy would be helpful,

specifically on the conditions that lead to higher rates of mis-

reporting (i.e., characteristics of students, questions, and

interview setting). Second, student self-reporting has found

its way into many state and local accountability systems. It

is important for researchers and practitioners to learn much

more about the types of self-report information that is more

or less accurate. While student surveys proliferate, the field

has little evidence to address whether students can self-

report accurately, beyond overall GPA and ACT/SAT scores.

The results can provide guidance to states and localities con-

sidering more reliance on student self-reporting.

Methodology and Descriptive Results

The HSLS:09 is a nationally representative longitudinal

study of >23,000 9th graders in U.S secondary schools. The

HSLS:09 base-year data collection took place in the 2009–

2010 school year and included surveys of students, parents

teachers, school counselors, and school administrators. The

first follow-up of HSLS:09 took place in 2012, when most

sample members were in 11th grade, and it included sur-

veys of students, parents, school counselors, and school

administrators. The 2013 update (designed to collect infor-

mation on the cohort’s postsecondary plans and choices)

occurred in the last half of 2013 and included surveys of

students and parents. Finally, high school transcripts were

collected in the 2013–2014 academic year. At each wave of

the study, surveys were conducted electronically (self-

administered), by phone, and in person via computer-

assisted interviewing methods. The content of the surveys

and the transcripts is quite extensive; for further informa-

tion, see Ingels et al. (2015).

We examined responses to questions about the grades that

students received in Algebra I, whether they enrolled in

Algebra I, and what other math courses they enrolled in. We

focused on math for the following reason. In sorting through

the extremely large number of courses taken by students in

the HSLS:09 data set, it was clear that math course naming

conventions (e.g., Algebra I, Geometry) are simple and stan-

dardized across schools. HSLS:09 does include data on sci-

ence and English language arts course taking. However,

science course titles were far less standardized, which we

believed introduced a high risk of misidentifying inaccurate

respondents. English course titles across the data set were in

a very simple sequence corresponding to the student’s grade

(English Language Arts I—9th grade, English Language

Arts II—10th grade, etc.). We therefore felt that matching

rates in English courses could reflect the simple ordering of

courses rather than something systematic about students’

ability to report their actual English courses.

The questions and response options are listed in Table 1.

Note that we report results using three different waves of the

survey: the base year, during the 9th grade; the first follow-

up, when most students were in 11th grade; and the second

follow-up, when most students were in the 12th grade.

Mathematics Course-Taking Accuracy

To correctly identify student responses as matches, we

adopt the following approach. First, we rely on course cod-

ing as conducted by the National Center for Education

TABLE 1

HSLS:09 Survey Questions and Response Options for Mathematics Courses

Question Response options Survey wave HSLS:09 variable

Are you currently taking a math course this fall? Yes/no 9th grade (2009) S1MFALL09

What math course(s) are you currently taking? (Check all that

apply.)

9th grade (2009)

Algebra I including IA and IB Check all that apply S1ALG1M09

Geometry Check all that apply S1GEOM09

Algebra II Check all that apply S1ALG2M09

Trigonometry Check all that apply S1TRIGM09

Review or Remedial Math including Basic, Business,

Consumer, Functional or General math

Check all that apply S1REVM09

Integrated Math I Check all that apply S1INTGM09

Statistics or Probability Check all that apply S1STATSM09

Integrated Math II or above Check all that apply S1INTGM209

Pre-algebra Check all that apply S1PREALGM09

Analytic Geometry Check all that apply S1ANGEOM09

Other advanced math course such as pre-calculus or calculus Check all that apply S1ADVM09

Other math course Check all that apply S1OTHM09

What grade were you in when you took Algebra I? [(If you have

taken it more than once, answer for your most recent course. If

you are currently taking Algebra I, choose your current grade.)

/ (If you have taken it more than once, answer for your most

recent course.)]

1 = 8th grade or earlier 11th grade (2011) S2ALG1WHEN

2 = 9th grade

3 = 10th grade

4 = 11th grade

5 = 12th grade

6 = You have not taken

Algebra I yet

What was your final grade in Algebra I? 1 = A (between 90-100) 11th grade (2011) S2ALG1GRADE

2 = B (between 80-89)

3 = C (between 70-79)

4 = D (between 60-69)

5 = Below D (anything

less than 60)

6 = Your class was not

graded

7 = You haven’t

completed the course

yet

[Are you currently/Were you] taking a math course [during the

spring term of 2012?]

Yes/no 11th grade (2011) S2MSPR12

What math course or courses [are you currently taking/were you

taking during the spring term of 2012]?

11th grade (2011)

Pre-algebra Yes/no S2PREALGM12

Algebra I, 1A or 1B Yes/no S2ALG1M12

Algebra II Yes/no S2ALG2M12

Algebra III Yes/no S2ALG3M12

Geometry Yes/no S2GEOM12

Analytic Geometry Yes/no S2ANGEOM12

Trigonometry Yes/no S2TRIGM12

Pre-calculus or Analysis and Functions Yes/no S2PRECALC12

Advanced Placement (AP) Calculus AB or BC Yes/no S2APCALC12

Other Calculus Yes/no S2CALC12

Advanced Placement (AP) Statistics Yes/no S2APSTAT12

(continued)

Question Response options Survey wave HSLS:09 variable

Other Statistics or Probability Yes/no S2STAT12

Integrated Math I Yes/no S2INTGM112

Integrated Math II Yes/no S2INTGM212

Integrated Math III or above Yes/no S2INTGM312

International Baccalaureate (IB) mathematics standard level Yes/no S2IBMATHSTD12

International Baccalaureate (IB) mathematics higher level Yes/no S2IBMATHHI12

Business, Consumer, General, Applied, Technical, Functional,

or Review math

Yes/no S2REVIEWM12

Other math course Yes/no S2OTHM12

[Did [you/he/she] take/[Have/Has][you/your teenager] taken]

any high school courses for college credit [when [you/he/she]

[were/was] in high school] including AP courses, IB courses,

and other courses for college credit? [Include any courses that

[you/he/she] [are/is] taking now.]

Yes/no 12th grade (2013) S3ANYCLGCRED

Which of the following types of courses for college credit [did

[you/he/she] take/[have/has] [you/he/she] taken[ when [you/he/

she] [were/was] in high school]?

12th grade (2013)

Advanced Placement (AP) courses Yes/no S3AP

International Baccalaureate (IB) courses Yes/no S3IB

In which of the following subjects [did [you/he/she] take/[have/

has] [you/he/she] taken] AP courses?

12th grade (2013)

Math Yes/no S3APMATH

Science Yes/no S3APSCIENCE

Another subject S3APOTHER

In which of the following subjects [did [you/he/she] take/[have/

has] [you/he/she] taken] IB courses?

12th grade (2013)

Math Yes/no S3IBMATH

Science Yes/no S3IBSCIENCE

Another subject Yes/no S3IBOTHER

Source. Ingels et al. (2015).

TABLE 1 (CONTINUED)

Statistics, in which coders used high school transcripts and

high school course catalogs to assign individual HSLS:09

courses a School Courses for the Exchange of Data (SCED)

code. SCED is a common classification system for second-

ary school courses and is updated and maintained by a work-

ing group of state and local education agency representatives

who receive suggestions and assistance from a wide network

of subject matter experts at the national, state, and local lev-

els. In any validity study, what is used for validation is

assumed to be correct, and we assume that the course coding

was done correctly and that it accurately reflects what

courses a student actually took.

Second, we distinguish between general course titles and

specific course titles as listed on the student survey. Terms

for general course titles, such as “Geometry,” could refer to

the specific SCED course title Geometry or to a wide num-

ber of courses listed on the SCED, such as Analytic

Geometry, Informal Geometry, or Principles of Algebra and

Geometry. Students could easily use “Geometry” as a short-

hand reference to any of these courses. Conversely, specific

course titles, such as Algebra II, refer to specific courses

listed on the SCED, and students should have the ability to

distinguish such courses from other courses that contain

algebra content, such as Principles of Algebra and Geometry.

Third, we use the following set of criteria for classifying

a student response as correct: (1) For courses on the student

interview that have specific course titles, such as Algebra I,

student responses are classified as correct only if the corre-

sponding SCED course title appears on the transcript. (2)

For responses on the student interview that refer to a group

of related courses, such as Statistics or Probability, student

responses are classified as correct if they have taken at least

one course in the group. In this example, any course with

“Statistics” or “Probability” in the SCED title would be

coded as a correct response. (3) Finally, we code student

responses in three ways for general course titles:

 Responses are correct only for the specific course

title, based on the full sample of students. For exam-

ple, a student who checks Geometry on the student

interview is coded as having a correct response only

if he or she took a course with the specific SCED title

of Geometry. In Table 2, this group is reported as a

match type of “specific” and “full” sample.

 Responses are correct for only the course title, after

restricting the sample by removing any student from

the analytic data set who took a course with a related

title. For example, any student who took a course

other than Geometry that contains “Geometry” in the

title, such as Informal Geometry, is dropped from the

analysis. This provides a measure of accuracy for stu-

dents who cannot be confused about what course

Geometry refers to, as these students took either no

Geometry course or only the course with the specific

TABLE 2

Percentage of Responses Matching Between Student Transcripts and Survey Reponses

Survey wave: Course Match type Sample Taken on transcript Taken on survey Match

9th grade (2009)

Pre-algebra Specific Full 3.5 4.9 94.7

Algebra I Specific Full 58.6 50.6 82.3

Algebra II Specific Full 4.8 6.4 96.1

Geometry Specific Full 24.3 23.9 94.7

Specific Restricted 24.9 23.5 95.5

General Full 26.7 23.9 94.1

Analytic Geometry Specific Full 0.1 0.2 99.8

Trigonometry Specific Full 0.1 0.4 99.6

Specific Restricted 0.1 0.2 99.7

General Full 0.5 0.4 99.5

Integrated Math I, II or above Group Full 4.7 3.0 95.6

Statistics or Probability Group Full 0.0 0.3 99.7

11th grade (2011)

Pre-algebra Specific Full 0.4 1.4 98.4

Algebra I, 1A or 1B Specific Full 5.4 5.7 94.0

Algebra II Specific Full 33.3 36.3 86.6

Algebra III Specific Full 1.7 3.9 95.5

Geometry Specific Full 13.5 14.7 92.6

Specific Restricted 13.9 14.0 93.6

General Full 16.3 14.7 92.1

Analytic Geometry Specific Full 0.2 0.5 99.4

Specific Restricted 0.2 0.5 99.5

General Full 0.8 0.5 98.8

Trigonometry Specific Full 2.4 9.4 91.8

Specific Restricted 2.5 5.8 95.5

General Full 9.7 9.4 92.5

Pre-calculus or Analysis & Functions Specific Full 18.3 19.3 94.8

Specific Restricted 18.7 18.7 95.8

General Full 20.5 19.3 94.8

AP Calculus AB or BC Group Full 2.5 2.9 99.3

AP Statistics Specific Full 0.9 1.1 99.6

Integrated Math I, II, III or above Group Full 3.7 3.2 95.8

IB mathematics standard level Specific Full 0.3 0.8 99.2

IB mathematics higher level Specific Full 0.4 0.1 99.5

12th grade (2013)

AP math courses Group Full 14.1 16.5 95.6

IB math courses Group Full 1.4 1.3 99.4

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W1STUTRN for 9th-grade courses, W3W2STUTRN for 11th-grade courses, and W3STUDENTTR for 12th-grade

courses.

SCED title of Geometry. In Table 2, this group is

reported as a match type of “specific” and “restricted”

sample.

 Responses are correct for any general course title that

is related to the student response. Here, any SCED

course title containing “Geometry” is coded as a cor-

rect response for a student who chooses Geometry on

the student interview. In Table 2, this group is reported

as a match type of “general” and “full” sample.

Fourth, some interview responses are so vague that coding

them correctly from a student’s point of view is difficult, if

not impossible. These are not included in our analysis; some

examples are “Review or remedial math including basic,

business, consumer, functional or general math” and “Other

advanced math course such as pre-calculus or calculus.”

To get a sense of the extent of possible misreporting, we

examined match rates for the math courses that students took

(see Table 2). The tables show what percentage of students

overall took a specific course (based on transcript data), the

percentage taking a specific course (based on the student sur-

vey), and the percentage of respondents for whom the tran-

script and self-report response match. For example, transcript

data indicate that 3.5% of the sample took pre-algebra, while

the student survey data indicate that 4.9% took pre-algebra.

Comparing transcripts to student self-reports of course taking

reveals that the pre-algebra self-report matches the transcript

record of courses taken for 94.7% of the sample.

Most courses have matching rates well above 90%, even

though most of the courses enrolled very few sample mem-

bers. However, for the courses that enrolled the most sample

members, such as Algebra I in the 9th grade and Algebra II

in the 11th grade, the match rates are somewhat lower—82%

and 87%, respectively. So, while the matching rates are high

across all the courses that we examined, the most commonly

taken courses show somewhat higher rates of mismatch.

Overall, students appear to be able to correctly report which

courses they have taken in high school.

Algebra I Reporting Accuracy

Students appear to accurately report which courses they

have taken, perhaps because courses taken are relatively dis-

tinct items that should be easy to recall and report accurately.

We next look at two items that should be more difficult to

report accurately: the year that Algebra I was taken and the

final grade received in the Algebra I course.

In Table 3, the bolded numbers on the diagonal reflect the

percentage of cases with matching transcript and student

survey data.

Percentages that are not bolded reflect mis-

matches. For example, for students with transcripts indicat-

ing that they took Algebra I in the 9th grade, 81% accurately

reported as much on the student survey. Across Table 3, for

transcripts indicating that Algebra I was taken at the 10th

grade or later, students reported it much less accurately, at

<50%. It is also notable that students who incorrectly

reported when they took Algebra I most often reported that

they took the course in the 9th grade.

Determining whether students accurately reported their

final grade in Algebra I is complicated by the lack of a final

grade accounting for every term that the student took the

course. Algebra I grades are reported for more than one term

on 57% of student transcripts that had at least one term of

Algebra I reported. Instead of reporting a single final grade,

schools report grades for four quarters, three trimesters, or

two semesters such that many students have multiple Algebra

I grades on their transcripts. We used the final grade reported

on the transcript for matching grades in Algebra I. We

believe that students are most likely to accurately recall their

final grades as opposed to grades that they receiving during

a semester or trimester.

In Table 4, the bolded percentages on the diagonal again

reflect the percentage of students whose survey response

matches what is recorded on their transcript. Students with

better grades in Algebra I tend to report more accurately. Of

the students who received an A in Algebra I, 83% correctly

reported receiving an A on the student survey. The accuracy

TABLE 3

Weighted Percentage of Student Interviews and Transcripts Matching on Grade When Algebra I Taken, by Grade

Students reporting Algebra I taken in

Transcript indicates taken in Overall 8th 9th 10th 11th 12th Not yet n

8th grade or earlier 31.1 80.5 15.0 2.3 0.6 0.0 1.6 5,860

9th grade 48.2 10.0 84.0 4.4 1.1 0.0 0.5 8,910

10th grade 10.3 7.9 47.7 40.0 3.0 0.1 0.9 1,600

11th grade 4.7 10.3 49.0 10.3 28.6 0.0 1.4 770

12th grade 0.1 0.0 69.8 0.0 21.4 6.1 2.6 20

Had not taken Algebra I yet 5.6 22.1 51.8 12.1 3.4 0.0 10.6 940

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W2STUTR. Bold indicates the percentage of cases with matching transcript and student survey data (percentages that

are not bolded reflect mismatches).

TABLE 4

Weighted Percentage of Student Interviews and Transcripts

Matching on Grade in Algebra I

Students reporting Algebra I grade

Transcript

final grade Overall A B C D Below D n

A 19.0 83.3 14.6 2.0 0.1 0.0 2,730

B 30.3 37.8 52.6 8.1 1.3 0.2 4,080

C 29.5 9.2 47.7 35.6 6.2 1.4 3,330

D 19.1 3.5 23.7 46.0 19.6 7.2 1,980

Below D 2.0 2.1 11.3 29.9 31.9 24.9 200

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W2STUTR. Bold indicates the per-

centage of students whose survey response matches what is recorded on the

transcript (mismatch otherwise).

rates for other transcript grades (B, C, D, below D) decline

fairly dramatically, ranging from 20% to 53%. Furthermore

and not surprising, students who misreport tend to inflate

their grades. For example, of the students who had a tran-

script grade report of B in Algebra I, 38% reported receiving

an A. This pattern is evident among students who received

C’s on their transcripts as well. When misreporting, students

tended to inflate their grades.

Multivariate Models

Our descriptive analyses demonstrate that (1) students are

reasonably good reporters of their course taking overall,

although the most commonly enrolled in courses (Algebra I

and Algebra II) show higher rates of misreporting; (2) stu-

dents who take Algebra I in 10th grade and beyond tend to

report inaccurately on the year when they took it; and (3)

students with lower grades in Algebra I tend to inaccurately

report their grades and, when they do misreport, seem to

inflate their academic performance. These results raise the

question why misreporting occurs on student surveys. To

address this question, we estimate two logistic regression

models predicting correct matches between student tran-

scripts and interviews on (1) what grade the student was in

when she or he took algebra and (2) the final grade received

in the Algebra I course, as reported on the transcript. We

used the same set of independent variables for each model.

Independent Variables

Descriptive statistics are presented in Tables 5 and 6.

Cognitive and academic ability. Student ability, in terms of

their cognitive ability and their performance in school, could

explain their reporting accuracy. We measured cognitive

ability with a math assessment

and academic ability with

the student’s overall GPA as reported on the transcript.

We also include dummy variables for the grade level in

which the student took Algebra I. It may be the case that

students who took the course most recently would be best

able to report accurately.

Conscientiousness on the survey task. Responding to sur-

veys is an effortful task. As discussed previously, Tourangeau

et al. (2000) outline a lengthy four-step cognitive process

that respondents go through before responding to a survey

item. At each point, error can be introduced. Following Hitt

et al. (2016), we include the percentage of items that a

respondent skips as a measure of conscientiousness on the

survey task, where students who answered <90% of items

are coded as 1 (0, otherwise). Approximately 6.5% of stu-

dents answered <90% of survey items.

Interview mode. Students were first surveyed in school via

computer-based self-administration. If the student failed to

participate in the school survey, she or he was then contacted

first for a telephone interview and then for an in-person

interview. Because the latter two modes introduce a human

interviewer into the process, students may react to social

desirability effects. Social desirability involves the inter-

viewee providing an answer that makes one “look good.” In

self-administered surveys, social desirability should be min-

imal (for a review, see Weisberg, 2005). The presence of an

interviewer (telephone or in person), we expect, may result

in social desirability–driven inaccuracy.

English language proficiency. As described by Tourangeau

et al. (2000), basic comprehension is required to process

TABLE 5

Descriptives for Independent Variables

Variable M SD Min Max

Took in 9th grade 0.67 0.47 0 1

Took in 10th grade 0.14 0.35 0 1

Took in 11th or 12th 0.06 0.24 0 1

Item skipper 0.06 0.23 0 1

Grade point average 2.66 0.76 0 4

Math cognitive ability 49.75 9.25 26.54 84.91

Mode: self, out of school 0.10 0.30 0 1

Mode: telephone 0.06 0.24 0 1

Mode: in person 0.06 0.24 0 1

English language learner status 0.03 0.16 0 1

Socioeconomic status −0.09 0.70 −1.75 2.28

Female 0.51 0.50 0 1

Asian 0.03 0.17 0 1

Black 0.14 0.35 0 1

Latino 0.23 0.42 0 1

Other race/ethnicity 0.09 0.29 0 1

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W2STUTR.

survey questions and report accurately. We suspect that

English language learners (ELLs) may have more difficulty

understanding questions, which may influence their ability

to report accurately. However, the survey questions that we

examined are fairly straightforward, and status as a limited

English speaker may not influence reporting accuracy as

much as it would if the survey questions under examination

were more complex. The ELL measure used in these models

is reported on the student transcript.

Model Results

Table 7 shows the results of the two logistic regression

models. For each model, we present the coefficients and the

discrete change in the probability of a correct match, for the

given amount of change in the independent variable, as indi-

cated in the last column of the table.

Timing of the event is a strong predictor of accuracy for

when the course was taken but not for the overall letter

grade. Students who took Algebra I in later grades were 11 to

14 percentage points more accurate than students who took

it in the 8th grade.

Our measure of survey engagement also showed mixed

results in terms of its effect on accuracy. Item skipping had

no statistically significant effect on letter grade accuracy, but

item skipping did have an effect on when the course was

taken. Students who skipped a large proportion of items

were less accurate in reporting when they took Algebra I, by

6 percentage points.

Cognitive ability is a strong predictor of accurate report-

ing with the student’s transcript-reported GPA and math

assessment score. The results are statistically and substan-

tively significant. If a student’s transcript-reported GPA

increases by 1 grade point, the probability of accurately

reporting the Algebra I grade increases by 18 percentage

points, and the probability of accurately reporting when

Algebra I was taken increases by 5 percentage points. The

corresponding changes for a 1-SD increase in math assess-

ment score are 1 and 4 percentage points, respectively.

Mode effects of the interview are generally negative, as

expected, given previous findings in the literature indicating

that self-reports tend to be the most accurate when assessed

without the presence of a human interviewer or others pres-

ent. Taking the survey online out of school slightly decreases

accuracy for when taken, and this could be due to the pres-

ence of family members and friends during survey adminis-

tration. Conducting the survey by telephone with a human

interviewer has a slight negative effect on accuracy, although

not statistically significant. Conducting the interview in per-

son reduces accurate reporting on grade received by 8 per-

centage points, but this is not statistically significant.

We note that these mode effects must be interpreted with

caution, because the design of the HSLS:09 changed the

interview mode as the number of contacts increased. That is,

those students who ended up with the in-person human inter-

viewer were surveyed via this mode because of previous

failed attempts to survey them with another mode. While

many of the characteristics correlated with refusal to

TABLE 6

Correlations for Independent Variables

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

1. Took in 9th grade 1.00

2. Took in 10th grade −.58 1.00

3. Took in 11th or 12th −.36 −.10 1.00

4. Item skipper .01 .00 .06 1.00

5. Grade point average .10 −.25 −.17 −.12 1.00

6. Math cognitive ability −.02 −.20 −.12 −.13 .54 1.00

7. Mode: self, out of school −.03 .01 .03 −.03 −.02 .00 1.00

8. Mode: telephone −.02 .04 .04 −.02 −.13 −.06 −.08 1.00

9. Mode: in person −.05 .04 .07 .02 −.17 −.11 −.08 −.06 1.00

10. English language learner status −.03 .08 .00 .03 −.07 −.08 .00 .04 .03 1.00

11. Socioeconomic status .03 −.13 −.09 −.06 .33 .34 −.02 −.05 −.10 −.12 1.00

12. Female .02 −.04 −.02 −.01 .16 −.01 .04 −.02 −.03 .00 .01 1.00

13. Asian −.02 −.03 −.01 .00 .10 .12 .00 −.01 −.01 .09 .04 −.01 1.00

14. Black .03 .01 −.02 .08 −.16 −.18 .01 −.01 .07 −.04 −.12 .04 −.07 1.00

15. Latino −.05 .07 .07 .04 −.19 −.13 .02 .07 .05 .18 −.28 .00 −.09 −.22 1.00

16. Other race / ethnicity .01 .01 .00 −.04 −.07 −.03 .01 .01 .02 −.05 −.01 −.01 −.06 −.13 −.18

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W2STUTR.

respond, such as academic ability and socioeconomic status,

are included in the models, it is possible that the mode effects

are picking up unobserved characteristics of students who

cooperate with surveys only after many attempts at refusal

conversion.

Perhaps the most surprising finding in the table is the

positive effect of ELL status, with ELLs more likely to

report accurately for grade received (18 percentage points)

and when taken (6 percentage points; not statistically signifi-

cant). After controlling for cognitive ability and survey

TABLE 7

Factors Affecting Correct Response for Algebra I Questions: Grade Received and When Taken

Letter grade received Grade when taken

B (SE) ΔP(Y = 1) B (SE) ΔP(Y = 1) ΔX

Took in 9th grade

−0.066 −0.02 1.083 0.14 0 to 1

(0.109) (0.132)**

Took in 10th grade

−0.034 −0.01 1.129 0.14 0 to 1

(0.172) (0.215)**

Took in 11th or 12th

0.181 0.04 0.768 0.11 0 to 1

(0.217) (0.211)**

Item skipper 0.018 0.00 −0.479 −0.06 0 to 1

(0.186) (0.210)*

Grade point average 0.749 0.18 0.413 0.05 1 grade point

(0.084)** (0.081)**

Math cognitive ability 0.005 0.01 0.038 0.04 1 SD

(0.006) (0.007)**

Mode: self, out of school

0.015 0.00 −0.163 −0.02 0 to 1

(0.151) (0.176)

Mode: telephone

0.044 0.01 −0.231 −0.03 0 to 1

(0.180) (0.168)

Mode: in person

−0.336 −0.08 −0.227 −0.03 0 to 1

(0.189) (0.213)

English language learner status 0.784 0.18 0.579 0.06 0 to 1

(0.328)* (0.380)

Socioeconomic status 0.012 0.00 0.069 0.01 1 SD

(0.055) (0.070)

Female 0.006 0.00 0.091 0.01 0 to 1

(0.076) (0.101)

Asian

0.178 0.04 −0.011 0.00 0 to 1

(0.219) (0.239)

Black

−0.238 −0.05 −0.139 −0.02 0 to 1

(0.126) (0.134)

Latino

−0.170 −0.04 0.184 0.02 0 to 1

(0.125) (0.164)

Other race/ethnicity

0.007 0.00 0.020 0.00 0 to 1

(0.100) (0.147)

Intercept −2.294 −1.867

(0.315)** (0.376)**

n 14,750 18,680

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W2STUTR.

Reference category is “took in 8th grade.”

Reference category is “mode: self, in school.”

Reference category is “white.”

*p < .05. **p < .01.

Rosen et al.

engagement, socioeconomic status had no effect on accurate

reporting.

Discussion

It is reassuring that, in general, reports of course taking

seem to contain manageable misreporting. In courses with

large cross sections of students (Algebra I and II), match

rates are lower, but students seem to be reasonably good

reporters of their courses. The slightly lower match rate in

courses with broad enrollment (Algebra I and II) is likely

driven by the presence of a broader cross section of students

in the sample (i.e., higher numbers of lower-ability students).

Unfortunately, reports on more sensitive questions—when

Algebra I was taken and what grade was received in it—con-

tain much higher rates of error.

While misreporting can affect descriptive statistics, one

major question is whether misreporting has an effect on mul-

tivariate analyses. That is, does misreporting affect the rela-

tionship between, for example, self-reported grades and

student outcomes? We estimated a simple model predicting

graduation from high school, using item skipping, GPA,

math cognitive ability, and demographics. We also included

the letter grade and grade taken for Algebra I from the tran-

script, as well as a variable measuring the discrepancy

between the transcript report and the self-report, by subtract-

ing the self-report response from the transcript report. A unit

change on this variable indicates a one-unit discrepancy

between the two sources of information (e.g., a student

reporting an A while the transcript reports a B).

The results of these models are reported in Table 8. The

discrepancy variable is statistically significant only in the

letter grade–taken model, and it suggests that errors in self-

reports are correlated with the probability of graduating

from high school. Students who overreport their grades by 1

letter grade have a probability of graduation 1 percentage

point lower than students who accurately report; overreport-

ing by 2 letter grades results in a decrease in probability of

graduating by almost 3 percentage points. Thus, it appears

that error in student self-reports is problematic for descrip-

tive statistics, as might be used by school districts, and for

academic researchers estimating multivariate models.

Similar to previous findings in the literature, our multi-

variate models confirm that some student characteristics,

primarily academic and cognitive ability, are major influenc-

ers of inaccurate survey reporting among students. Higher-

ability students report more accurately than lower ability

students, and this is consistent across both of our models.

Since the survey reporting task involves a number of traits

and attributes (i.e., conscientiousness, persistence, aptitude)

that correlate with cognition and performance school, it is

not surprising that cognitive ability and classroom perfor-

mance would predict accuracy in survey reports. However,

the magnitude of the effect is notable. Higher-GPA students

are far more accurate reporters, as demonstrated in our mul-

tivariate model predicting accuracy in reporting Algebra I

grades.

ELLs did not behave as we predicted, nor perhaps as the

Tourangeau et al. (2000) model would predict. ELLs were

more likely to report their grades accurately than non-ELLs

but not more likely to report the grade when they took

Algebra I. Theorizing from the Tourangeau et al. model, we

expected ELLs to be less likely to report accurately due to

potential question comprehension problems. Rather, we

found the opposite. Unfortunately, our data do not allow us

to investigate this further, and we do suggest that future

research examine the intersection of cultural and social

TABLE 8

Predicting High School Graduation With Algebra I Responses

1 2

Transcript Algebra I: letter grade 0.210

(0.147)

Transcript Algebra I: when taken −0.122

(0.122)

Transcript minus self-report −0.252 0.036

(0.103)* (0.121)

Item skipper −0.126 −0.306

(0.402) (0.339)

Grade point average 2.059 1.797

(0.191)** (0.149)**

Math cognitive ability 0.018 0.016

(0.014) (0.013)

Item skipper −0.126 −0.306

(0.402) (0.339)

English language learner status −0.470 −0.420

(0.538) (0.453)

Socioeconomic status 0.362 0.352

(0.153)* (0.144)*

Female −0.016 −0.028

(0.189) (0.164)

Asian

−0.728 −0.490

(0.598) (0.498)

Black

0.033 0.106

(0.204) (0.181)

Latino

0.307 0.364

(0.222) (0.183)*

Other race/ethnicity −0.219 −0.222

(0.354) (0.314)

Intercept −3.157 −1.765

(0.916)** (0.725)*

n 12,290 16,980

Source. Ingels et al. (2015).

Note. Estimates are weighted with W3W2STUTR. Values are presented

as B (SE).

Reference category is White.

*p < .05. **p < .01.

Accuracy of Student Self-Reports

forces, language ability, and measurement errors on student

surveys.

Inaccurate reporting by students does not appear to be ran-

dom. In fact, students in our sample may even be making

rational choices when they misreport. By application of the

Tourangeau et al. (2000) model to the reporting of grades, the

response phase gives students a chance to decide if (and how)

they should report potentially embarrassing information. A

low final grade might be embarrassing for a student to report.

When a student erroneously reports a grade, he or she does

tend to inflate that grade but perhaps only slightly. For exam-

ple, C students were more likely to report receiving a B

(47.7%) than an A (9.2%). D students were more likely to

report a C (46.0%) than a B (23.7%) or an A (3.5%). When

the student decides to misreport during the response phase,

she or he may make a rational choice to misreport reasonably.

Our data do not allow us to investigate this further, but it

could be the case that rational decision making goes into the

decision to misreport. If this is in fact the case, survey meth-

odologists will need to develop and test methods of control.

For example, prompts in the instrument indicating that

answers may be cross-checked with transcripts could encour-

age more accuracy. If students are acting rationally, perhaps

they would respond to prompts with more accurate reports.

Systematic patterns of misreporting may be evident in

other ways. Students who misreported when they took

Algebra I were most likely to report taking the course in 9th

grade. This could reveal something systematic about misre-

porting, such as confusion about the survey question or mis-

chief on behalf of the respondent. We investigated the

transcript-recorded courses of students who inaccurately

reported taking Algebra I in 9th grade. In 37% of these cases,

Algebra I was taken in multiple years (9th and 10th grades)

but reported for 9th grade, even when the survey question

asked respondents to report the “most recent grade” that they

took the course. This pattern may reflect respondent confu-

sion over the question wording. In another 37% of cases that

misreported taking Algebra I in 9th grade, the transcripts

show the course being completed in the 8th grade and the

student enrolled in Geometry or Integrated Math in 9th grade.

The remaining 25% of cases have transcripts that show 9th-

grade enrollment in Pre-Algebra or some other unclassified

math course that may be a lower level than what is typical for

9th grade. This pattern may reflect social desirability bias.

The literature indicates that higher-performing students

are fairly accurate self-reporters of distinctive and easily

recalled information, such as GPA. However, lower-per-

forming students are often inaccurate when they self-report

school performance measures. We sought to determine if

this pattern was evident in factual questions that should be

easier to recall. As with school performance measures, we

found lower-performing students to be less accurate than

higher-performing students when they reported on the grade

level in which they were enrolled in Algebra I. Remembering

whether specific courses were taken should be relatively

easy, as opposed to other information about a specific course.

Our data show that students can accurately recall and report

whether they took Algebra I, but they are much less able to

accurately report when they took the course or, especially,

how they performed in it. This pattern of results reflects the

major obstacle to researchers and administrators seeking to

use student self-reported data in their work. Relatively dis-

tinctive and easily recalled information lends itself to accu-

rate reporting. But more frequent and mundane events are

less likely to be encoded in memory, withdrawn from mem-

ory, and accurately reported. Yet it is often these more fre-

quent and mundane behaviors that we wish to have to use in

our work.

These results lead us to question the use of many types of

questions in student surveys. Many of these questions ask

students about frequent mundane events, sometime asking

them to report over periods of a year or more. It is vital that,

as a field, we establish a firmer research base for the use of

these questions before we can begin to use them in our

research and accountability efforts.

Notes

1. See http://tripoded.com/about-us-2/.

2. See http://coredistricts.org/.

3. See http://schools.nyc.gov/Accountability/tools/survey/default

.htm.

4. HSLS:09 did not collect transcripts from middle schools;

however, some high schools provided high school–level courses

taken in the 8th grade or earlier. For high school transcripts that

did not provide that information, we were able to identify students

who took Algebra I in the 8th grade or earlier when they took a

higher-level math course in the 9th grade (e.g., geometry, Algebra

II, trigonometry).

5. We also conducted an analysis to assess how sensitive match

rates were to different coding decisions. For students who had mul-

tiple Algebra I grades reported on their transcripts, we calculated an

average course grade and used that for matching. For example, if a

student’s transcript reported an A in Algebra I for the fall semester

and a B for the spring semester, these grades are recorded as 4.0 and

3.0, respectively, and we calculated the final grade as the average,

3.5. Coding in this manner resulted in an Algebra I grade match

rate of 48.9%, compared with 46.9% based on the final grade only.

6. HSLS:09 does not include a reading cognitive assessment.

7. The item response percentage was calculated as the number

of items to which a student responded, divided by the total number

of items that applied to that student.

References

Caskie, G. I. L., Sutton, M. C., & Eckhardt, A. G. (2014). Accuracy

of self-reported college GPA: Gender-moderated differences

by achievement level and academic self-efficacy. Journal of

College Student Development, 55(4), 385–390.

Cassady, J. C. (2001). Self-reported GPA and SAT: A methodolog-

ical note. Practical Assessment, Research & Evaluation, 7(12).

Retrieved from http://PAREonline.net/getvn.asp?v=7&n=12

Rosen et al.

Cole, J., & Gonyea, R. M. (2010). Accuracy of self-reported SAT

and ACT test scores: Implications for research. Research in

Higher Education, 51(4), 305–319.

Darolia, R. (2014, February). Working (and studying) day and

night: Heterogeneous effects of working on the academic per-

formance of full-time and part-time students. Economics of

Education Review, 38, 38–50.

Diseth, A., Meland, E., & Breidablik, H. J. (2014). Self-beliefs

among students: Grade level and gender differences in self-

esteem, self-efficacy and implicit theories of intelligence.

Learning and Individual Differences, 35, 1–8.

Duckworth, A. L., & Yeager, D. S. (2015). Measurement mat-

ters: Assessing personal qualities other than cognitive ability.

Educational Researcher, 44(4), 237–251.

Feldman, M., & Kubota, M. (2015). Hope, self-efficacy, optimism,

and academic achievement: Distinguishing constructs and lev-

els of specificity in predicting college grade-point average.

Learning and Individual Differences, 37, 210–216.

Ferguson, R. F. (2008). The TRIPOD Project framework.

Cambridge, MA: Harvard University.

Ferguson, R. F. (2012). Can student surveys measure teaching

quality? Phi Delta Kappan, 94(3), 24–28.

Guo, J., Marsh, H. W., Morin, A. J. S., Parker, P. D., & Kaur,

G. (2015). Directionality of the associations of high school

expectancy-value, aspirations, and attainment: A longitudi-

nal study. American Educational Research Journal, 52(2),

371–402.

Hitt, C., Trivitt, J., & Cheng, A. (2016). When you say noth-

ing at all: The predictive power of student effort on surveys.

Economics of Education Review, 52, 105–119.

Ingels, S. J., Pratt, D. J., Herget, D., Bryan, M., Fritch, L. B., Ottem,

R., . . .Wilson, D. (2015). High School Longitudinal Study

of 2009 (HSLS:09) 2013 update and high school transcript

data file documentation (Document No. NCES 2015-036).

Washington, DC: National Center for Education Statistics.

Krosnick, J. A., & Presser, S. (2010). Question and questionnaire

design. In P. V. Marsden, & J. D. Wright (Eds.), Handbook of

survey research (2nd ed., pp. 263–314). Bingley, UK: Emerald

Group.

Kuncel, N. R., Credé, M., & Thomas, L. L. (2005). The valid-

ity of self-reported grade point averages, class ranks, and test

scores: A meta-analysis and review of the literature. Review of

Educational Research, 75(1), 63–82.

Leaper, C., Farkas, T., & Brown, C. S. (2012). Adolescent girls’

experiences and gender-related beliefs in relation to their

motivation in math/science and English. Journal of Youth

Adolescence, 41(3), 268–282.

Maxwell, N. L., & Lopus, J. S. (1994). The Lake Wobegon effect

in student self-reported data. American Economic Review,

84(2), 201–205.

Mayer, R. E., Stull, A. T., Campbell, J., Almeroth, K., Bimber, B.,

Chun, D., . . . Knight, A. (2007). Overestimation bias in self-

reported SAT scores. Educational Psychology Review, 19(4),

443–454.

Meade, A. W., & Craig, S. B. (2012). Identifying careless responses

in survey data. Psychological Methods, 17, 437–455.

Niemi, R. G., & Smith, J. (2003). The accuracy of students’

reports of course taking in the 1994 National Assessment of

Educational Progress. Educational Measurement: Issues and

Practice, 22(1), 15–21.

Ratelle, C. F., & Duchesne, S. (2014). Trajectories of psychological

need satisfaction from early to late adolescence as a predictor of

adjustment in school. Contemporary Educational Psychology,

39(4), 388–400.

Robinson-Cimpian, J. (2014). Inaccurate estimation of disparities

due to mischievous responders: Several suggestions to assess

conclusions. Educational Researcher, 43(4), 171–185.

Sanchez, E., & Buddin, R. (2015). How accurate are self-reported

high school courses, course grades, and grade point average?

(ACT Working Paper Series No. WP-2015-03). Iowa City, IA:

American College Testing Program.

Sawyer, R., Laing, J., & Houston, M. (1988). Accuracy of self-

reported high school courses and grades of collegebound

students (ACT Research Report No. 88-1). Iowa City, IA:

American College Testing Program.

Schwartz, J. A., & Beaver, K. (2015). Making (up) the grade?

Estimating the genetic and environmental influences of discrep-

ancies between self-reported grades and official GPA scores.

Journal of Youth and Adolescence, 44(5), 1125–1138.

Teye, A. C., & Peaslee, L. (2015). Measuring educational out-

comes for at-risk children and youth: Issues with the validity of

self-reported data. Child & Youth Care Forum, 44(6), 853–873.

Tourangeau, R., Rips, L. J., & Rasinski, K. (2000). The psychology

of the survey response. Cambridge, UK: Cambridge University

Press.

Valiga, M. J. (1986). The accuracy of self-reported high school

course and grade information (ACT Research Report Series

No. 87-1). Iowa City, IA: American College Testing Program.

Yeager, D. S., Romero, C., Paunesku, D., Hulleman, C., Schneider,

B., Hinojosa, C., . . . Dweck, C. S. (2016). Using design think-

ing to make psychological interventions ready for scaling: The

case of the growth mindset during the transition to high school.

Journal of Educational Psychology, 108(3), 374–391.

Zimmerman, M. A., Caldwell, C. H., & Bernat, D. H. (2002).

Discrepancy between self-report and school-record grade point

average: Correlates with psychosocial outcomes among African

American adolescents. Journal of Applied Social Psychology,

32(1), 86–109.

Authors

JEFFREY A. ROSEN is a researcher at RTI International. His

research interests include data quality in sample surveys, social and

emotional learning, and educational interventions for traditionally

disadvantaged students.

STEPHEN R. PORTER is professor of higher education in the

Department of Educational Leadership, Policy, and Human

Development at North Carolina State University, where he teaches

courses in educational statistics and causal inference with observa-

tional data. His current research focuses on student success, with an

emphasis on quasi-experimental methods and survey methods, par-

ticularly the validity of college student survey questions.

JIM ROGERS is a senior manager of systems analysis and pro-

gramming at RTI International. His interests are in probability sur-

veys and data management.