ACT Research & Policy | Technical Brief | May 2020
A Case Study: ACT Section Retest Scores
and Superscores are Predictive of First-Term
Grades
Justine Radunzel, PhD, and Krista Mattern, PhD
Abstract
This study conducted in collaboration with a postsecondary institution highlights results
from a concurrent validity study of administering ACT
®
section tests to their entering
freshmen who previously took the ACT test in high school. Students’ ACT scores obtained
from section retesting were found to be as predictive of rst-term grade point average
(GPA) as scores obtained via traditional ACT testing. Additionally, ACT Superscores
that were computed across test administrations that included single-subject section test
events were found to be predictive of rst-term GPA, alone and in combination with high
school GPA. Moreover, the strength of this relationship did not signicantly dier from that
based on students’ most recent ACT Composite scores.
Introduction
ACT test scores are designed to measure students’ level of college readiness in key
core academic content areas (ACT, 2019). Postsecondary institutions use ACT scores in
combination with other measures such as high school GPA to help inform their admission
and placement decisions and to identify students most likely to struggle academically, be
at risk of dropping out, and benet from institutional services and supports (ACT, 2019;
Clinedinst, 2019; University of California Academic Senate, 2020). Numerous studies
have been conducted that provide validity evidence supporting the use of ACT scores for
these purposes (ACT, 2019; Mattern & Allen, 2016; Radunzel, 2017).
Beginning in September 2020, three new testing options will be available on national
ACT test dates: online testing, section retesting, and superscoring. Section retesting
(also referred to as modular testing or single-subject retesting) gives students the option
to retake one or more sections of the ACT test instead of having to take the full ACT test
again. Section retesting will initially only be available to students retesting online. The
new option of superscoring will allow students who have tested more than once to send
their ACT Superscore—the average of students’ highest scores in each subject from all
of their test attempts (including from section retests)—to postsecondary institutions of
their choice. The option of superscoring is in alignment with current admissions practices
and policies at many postsecondary institutions and allows students to demonstrate their
academic achievement most favorably for college applications and scholarships.
1
ACT, Inc. 2020
© by ACT, Inc. This work is licensed under a Creative Commons Attribution-Non
Commercial 4.0 International License.https://creativecommons.org/licenses/by-nc/4.0/
ACT.org/research
R1817
2
ACT Research & Policy | Technical Brief | May 2020
Over the last several years, ACT has been conducting research to examine the validity and
fairness of dierent scoring practices and options to help provide insights on how postsecondary
institutions might best make use of multiple ACT scores when students retest. Results from
the studies conducted to date support oering the new options of section retesting and
superscoring. First, in a large multi-institutional study (Mattern, Radunzel, Bertling, & Ho, 2018),
superscoring was found to be as predictive—if not more predictive—of rst-year GPA than
the other ACT Composite scoring methods examined, which included computing the average
Composite score across test administrations or using students’ most recent Composite score
or their highest Composite score; correlations ranged from 0.39 for the average to 0.41 for
superscoring. The study also found that rst-year GPA for students who tested more often was
underpredicted, but that when examining the prediction accuracy by the number of times tested,
superscoring resulted in the least amount of prediction error across the four scoring methods.
These results suggest that ACT subject scores do not have to come from a single test attempt
to be a valid indicator of students’ college readiness, supporting both superscoring and section
retesting.
The Mattern et al. (2018) study also explored the diversity implications for an admitted class of
using superscores as compared to the other three scoring methods to admit students. Despite
the fact that underserved students are less likely to retest (Harmston & Crouse, 2016), the
authors found that superscoring did not result in a less diverse admitted class as compared
to the other three scoring methods. In a subsequent study (Mattern & Radunzel, 2019), the
researchers found that superscoring did not exacerbate subgroup dierences for the national
ACT-tested population over those reported based on students’ most recent ACT scores.
Second, results from a randomized study of 4,000 students conducted in 2016 indicated that
the order in which the subject tests were administered did not impact student performance
(Andrews, 2019). More specically, the study found that students earned subject scores that
were similar regardless of the order in which the subject tests were taken. Given that ACT
scores were similar when taken rst as compared to the standard position in the full ACT test,
the ndings from this study support the option of oering section retesting where students will
not have to retake the entire ACT test but can focus their learning eorts on specic subject
areas of their choice. Despite concerns being raised that section retesting may lead to articially
inated scores, two recent studies (Mattern, Radunzel, & Andrews, 2019; Radunzel & Mattern,
2020) provide empirical evidence suggesting that this is actually not the case. In particular,
the results from these two studies demonstrate that students’ performance when retesting in a
single ACT subject area tends to be consistent with what would be expected based on typical
test-retest score gains from taking the entire ACT test.
While decades of research provide evidence that each individual ACT test is a valid and
reliable measure of students’ college readiness and related to college outcomes (ACT, 2019;
see chapters 10 and 11), there is a need to examine the predictive validity of section retest
scores. Moreover, given that the prior study on superscoring (Mattern et al., 2018) was based
only on full administrations of the ACT, it is of interest to investigate the relationship between
rst-year college outcomes and ACT Superscores that combine scores not only across full
test administrations but also across section retests. To address these topics, we conducted
a concurrent validity study in collaboration with a single four-year public university involving
students from their fall 2019 freshman cohort. In particular, the following two research questions
were examined in this case study:
3
ACT Research & Policy | Technical Brief | May 2020
1. Are section retest scores as predictive of rst-term grades as subject scores obtained from
taking the full ACT test?
2. Are ACT Superscores that include section retest scores, alone and in combination with high
school GPA, as predictive of rst-term grades as students’ most recent ACT Composite
scores?
Data and Methods
Study Sample
The participating institution is located in the western region of the country and has a highly
selective admissions policy. The incoming freshman class size for this institution is a little over
4,000 students. The institution accepts both ACT and SAT test scores; nearly 60% of students
submit ACT scores to the institution as part of the admissions process (U.S. Department of
Education, 2020).
The institution completed the following activities for the study: (a) assigned a study coordinator to
serve as the point of contact, (b) recruited rst-year, rst-time entering domestic college students
who had previously taken the full ACT test in high school to take a single subject ACT test during
the rst three weeks of the fall 2019 term, (c) administered and proctored single-subject ACT
tests in paper format in a secure manner and under standard testing conditions, (d) returned
the completed answer documents to ACT, and (e) submitted a data le of students’ rst-term
grades in January 2020. We aimed to recruit 50 students per ACT subject area (English, math,
reading, science) for a total of 200 students. Approximately a month before classes began, the
institution began inviting students to participate via email.
2
It became apparent early on that it
would be dicult to achieve this recruitment goal within the narrow study recruitment and testing
window. For this reason, students were allowed to test in multiple subject areas and to choose the
subject(s) they wanted to test in. Students who tested in multiple subject areas took each subject
test on dierent days to simulate a single-subject test experience.
Both the institution and study participants were compensated for participating in the study. The
institution received a monetary incentive for completing the study activities. Students received
a $50 gift card for each single-section test taken. To increase students’ motivation for testing,
students were informed that they would receive an additional $50 gift card if they met or exceeded
their most recent ACT subject score taken during their junior or senior year in high school.
Complete data was available for a total number of 118 students who had prior ACT scores.
3
The resulting sample size by subject area was 39 in English and math, 50 in reading, and 46
in science. The institution provided outcomes data on their entire 2019 freshman cohort. There
were 2,729 non-study participants who had taken the ACT test in high school. This sample of
ACT-tested nonparticipants was used for comparison purposes.
4
ACT Research & Policy | Technical Brief | May 2020
Measures
Outcome. The primary outcome was the fall-term college GPA, on a scale from 0.00 to 4.00. The
average rst-term GPA for the participant sample was 3.36 with a standard deviation of 0.52.
Predictors. The following academic achievement measures were included as predictors of
rst-term GPA: ACT section scores in English, math, reading, and science; the ACT Composite
score; and high school GPA. ACT section scores and the Composite score range from 1 to 36.
ACT subject scores were either the scores earned on the section tests that were administered
during the rst three weeks of college (labeled as section test) or the scores earned on students’
most recent traditional ACT testing event from their junior or senior year in high school (labeled
as full test-most recent).
Two dierent Composite scores were used; these included the most recent ACT Composite
score earned in high school and the ACT Superscore that was computed by combining the
highest subject scores across test administrations from a students’ sophomore, junior, and
senior year in high school, and their section test taken during the rst three weeks of college.
High school GPA was self-reported by students during the ACT registration process and is
based on their coursework taken in up to 23 specic courses in English, mathematics, social
studies, and science, and the grades earned in those courses. Prior studies have shown that
students report high school coursework and grades accurately relative to information provided
in their high school transcripts (Sanchez & Buddin, 2016).
Analysis
Means and percentages were used to describe the outcomes and student characteristics.
Pearson correlation coecients were computed between students’ achievement measures
and rst-term GPA. The formulas presented by Steiger (1980) were used to test whether two
correlations that involve a common outcome variable were signicantly dierent from one
another. Linear regression models were developed to predict rst-term GPA from students’
achievement measures. To address study objective 1, analyses were conducted separately for
each subject sample using the corresponding ACT section score. To address study objective
2, analyses were conducted for the full participant sample using ACT Composite scores. A
signicance level of .05 was used in this study.
Description of Samples
Table 1 provides descriptive information on ACT Composite score, high school GPA, and rst-
term GPA, by sample. On average, students who participated in the study had slightly higher
ACT Composite scores from high school and earned slightly higher rst-term GPAs than ACT-
tested nonparticipants from the institution. The average high school GPA tended to be more
comparable across the samples. The average time between when the ACT section test was
taken as a part of this study and when the full ACT was last taken in high school ranged from
13.6 months for the English sample to 15.1 months for the science sample.
As shown in Table A1 in the Appendix, there was representation in the study samples across
gender, race/ethnicity, socioeconomic status, and initial declared major category, though
dierences existed across samples. For example, the math sample had a higher percentage
of males and STEM majors than the other samples did (76.9% vs. 43.6% to 56.0% for males
and 79.5% vs. 54.1% to 67.4% for STEM majors). For each sample, the percentage of students
returning for the second term was high (ranging from 97.4% to 100.0%).
5
ACT Research & Policy | Technical Brief | May 2020
Table 1. Average ACT Composite Score, High School GPA, and First-term GPA by Sample
Variable
English
sample
Math
sample
Reading
sample
Science
sample
ACT-tested
nonparticipants
from institution
n Mean (SD) n Mean (SD) n Mean (SD) n Mean (SD) n Mean (SD)
ACT Composite
score – most
recent
39
29.6 (3.7)
39 30.4 (4.2) 50 31.2 (3.7) 46 30.6 (4.0) 2,729 27.7 (4.1)
Time between
ACT testing
(in months)
39
13.6 (4.2)
39 14.2 (4.0) 50 14.8 (4.5) 46 15.1 (4.8)
High school GPA 38
3.80 (0.19)
37 3.84 (0.16) 49 3.79 (0.24) 45 3.86 (0.15) 2,573 3.80 (0.22)
First-term GPA 39
3.37 (0.43)
39 3.35 (0.55) 50 3.45 (0.46) 46 3.41 (0.52) 2,729 3.16 (0.62)
Note. GPA = grade point average. SD = standard deviation. Time between ACT testing is the dierence in months
between when the ACT section test was taken as a part of this study and when the full ACT test was last taken in high
school (labeled as ACT Composite score – most recent).
Results
Section Scores
Table 2 provides average ACT scores for the two testing events by subject and sample, as well
as the correlations between the two scores. First, looking at the results in the table that are
labeled as full sample, we see that the average ACT scores on the section retest tended to be
relatively high, ranging from 27.9 in science to 29.7 in math. However, students’ scores on the
section retest completed during the rst three weeks of college tended to be lower than those
earned previously in high school when taking the full ACT (by 1.1 score points in math to 2.6
score points in English). This was seen even after omitting an extreme outlier from the math,
reading, and science samples.
4
For reference, a study by Harmston and Crouse (2016) found
that students rst testing as juniors demonstrated an average Composite score increase of 1.1
points by their nal ACT test. The magnitude of the average decline in scores may suggest that,
despite the incentives oered, students were not as motivated on the section retest as they
had been when taking the full ACT in high school to earn a college-reportable score. In fact,
compared to their latest test scores from high school, the percentages of students scoring lower
by more than two standard errors of measurement (SEM) on their section retest ranged from
16% in reading to 31% in English, which is higher than expected. If students’ true achievement
remained constant between the two time points, we would have expected fewer than 8% of
students to score lower by more than two SEM.
Because some students may not have been fully engaged on the section retest, we also present
results for a reduced sample that excludes students who experienced large score declines
(labeled as subsample in the table). Even though around 40% to 50% of students in the
subsample increased their subject scores on the section retakes, scores tended to be slightly
lower on the section retest than those earned previously in high school on the full ACT (by 0.2
point in math to 1.2 points in reading).
5
Compared to the full sample, the correlations between
6
ACT Research & Policy | Technical Brief | May 2020
the full test and section retest scores were higher for the subsample of students (ranging from
.77 to .94 vs. .65 to .73 for the full sample).
6
Given that students’ level of motivation could
signicantly impact the results of the study, it is unfortunate that we were unable to quantify and
therefore control for this factor in the analyses. This is a limitation of this study.
Table 2. Average ACT Scores and Correlations between ACT Scores by Subject and Sample
Subject n
Full test -
most recent Section test
Dierence in
ACT scores
Correlations between ACT
scores
Mean SD Mean SD Mean SD R
95%
Lower
95%
Upper
English
Full sample 39 31.0 4.6 28.4 4.7 -2.6 3.4 .73 .55 .85
Subsample 27 30.3 5.1 29.6 4.8 -0.7 1.7 .94 .88 .97
Math
Full sample 38 30.8 3.7 29.7 3.3 -1.1 2.9 .65 .42 .80
Subsample 31 30.4 3.8 30.2 3.3 -0.2 2.5 .77 .57 .88
Reading
Full sample 49 31.9 4.4 29.5 5.2 -2.4 3.9 .67 .49 .80
Subsample 42 31.7 4.6 30.5 4.8 -1.2 2.3 .88 .79 .94
Science
Full sample 45 29.6 4.7 27.9 5.0 -1.7 3.6 .72 .54 .84
Subsample 36 28.8 4.8 28.5 5.1 -0.4 2.4 .89 .79 .94
Note. SD = standard deviation. The full sample includes all students’ scores except the one outlier (see end note #4).
The subsample excludes students whose scores on section retesting decreased by more than 2 SEM compared to their
latest full test score from high school. All correlation coecients were signicantly dierent from 0 (p < .0001). Students
completed the entire ACT test during their junior or senior year in high school, while students completed the section
retest during the rst three weeks of their freshman year in college.
Figure 1 provides the predicted rst-term GPA as a function of ACT subject score by testing
event for the subsamples. Figure A1 in the Appendix shows the corresponding gures for the full
samples, and Table A2 in the Appendix provides the regression estimates by sample and testing
event. As shown in Figure 1 and Table A2, ACT single-section test scores were predictive
of rst-term GPA among the subsamples of students who may have been more engaged.
Moreover, in English and reading, the regression lines and parameter estimates associated with
section retest scores were similar to those estimated using students’ prior full ACT scores. In
math and science, the slope appeared to be slightly steeper for the section retest scores than
for the prior full ACT scores. However, the dierences in the predicted values between the two
scores were relatively small (at most 0.13 in math and at most 0.07 in science).
7
Moreover, as
shown in Table A2, the 95% condence intervals for the slopes associated with standardized
test scores overlapped between the two testing events, suggesting that the two slopes did not
signicantly dier from one another. Similar conclusions were reached for the full sample (see
Figure A1 and Table A2).
7
ACT Research & Policy | Technical Brief | May 2020
Figure 1. Predicted First-Term GPA by ACT Section Score and Testing Event for Subsamples
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT English score
Full test - most recent Section test
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT math score
Full test - most recent Section test
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT reading score
Full test - most recent Section test
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT science score
Full test - most recent Section test
Table 3 provides the correlation coecients between ACT subject scores and rst-term college
GPA by subject and sample. For the full sample, the correlation coecient between the section
retest score and rst-term college GPA was signicantly dierent from zero in each subject area
(as evidenced by the 95% condence intervals not including zero); the correlations ranged from
.31 in reading to .53 in math. This result was also seen for the subsample except in reading
where the correlation coecient was nearly signicantly dierent from zero (95% condence
interval = -.02 to .54); the correlations ranged from .29 in reading to .52 in math. The test-
criterion correlations for the section retest scores did not signicantly dier from those estimated
using students’ latest full ACT scores from high school (each p-value greater than .19).
Moreover, the estimated section retest correlations with rst-term GPA are in line with those
reported in other studies between ACT Composite score and rst-year college GPA (Mattern
et al., 2018; Sawyer, 2010; Westrick, Le, Robbins, Radunzel, & Schmidt, 2015). A possible
explanation for the correlation coecient and estimated slope being higher in math than in the
other subject areas is that there was a higher percentage of STEM majors among the sample
of students taking the math section test than among the other samples (79.5% vs. 58.0% to
67.4%, Table A1 in Appendix).
These results indicate that ACT scores earned in a modular setting are predictive of rst-term
GPA and provide a valid indicator of college readiness. In the next section, we examine the
validity of using ACT Superscores that include students’ section retest scores in its computation
to predict rst-term GPA, alone and in combination with high school GPA.
8
ACT Research & Policy | Technical Brief | May 2020
Table 3. Correlations (R) between ACT Scores and First-Term College GPA by Subject and
Sample
Subject n
Full test – most recent Section test
p-value
1
R
95%
lower
95%
upper R
95%
lower
95%
upper
English
Full sample 39 .34 .03 .59 .42 .12 .65 .486
Subsample 27 .44 .07 .70 .44 .07 .70 .954
Math
Full sample 38 .45 .15 .67 .53 .26 .73 .481
Subsample 31 .42 .07 .67 .52 .20 .74 .346
Reading
Full sample 49 .25 -.03 .50 .31 .04 .55 .576
Subsample 42 .29 -.02 .54 .29 -.02 .54 .990
Science
Full sample 45 .31 .02 .55 .36 .08 .59 .619
Subsample 36 .30 -.03 .57 .40 .08 .64 .198
Note. GPA = grade point average. 95% lower and 95% upper corresponds to the lower and upper limits for the 95%
condence intervals. The full sample includes all students’ scores except the one outlier (see endnote #4). The
subsample excludes students whose scores on section retesting decreased by more than 2 SEM compared to their
latest full test score from high school. Students completed the entire ACT during their junior or senior year in high
school, while students completed the section retest during the rst three weeks of their freshman year in college.
1
p-value corresponds to testing whether the two correlations signicantly diered from one another (Steiger, 1980).
ACT Superscores
Nearly three-fourths of all students in the study sample (74.6%) took the ACT test more than
once in high school; the average number of times tested with the full test was 2.3 times (not
including the section retest taken during the rst three weeks of college). ACT Superscores
were computed by combining the highest section scores across test administrations from a
students’ sophomore, junior, and senior year in high school and their single-section retest.
Superscores for nearly three-fourths of students were based solely on scores earned when
completing the ACT during high school. For the remaining 26.3% of the students, the score
earned on one of the section retakes was the highest section score across test administrations
and was utilized in computing the ACT Superscore.
8
As shown in Table 4, the average ACT
Superscore was slightly higher than the average of students’ most recent ACT Composite
scores from high school (by 0.7 point, 31.0 vs. 30.3, respectively). This dierence in average
scores between these two scoring methods is consistent with that reported in other studies
(Mattern & Radunzel, 2019; Mattern et al., 2018). The two Composite scores were highly
correlated (R = .97; 95% condence interval = .95, .98).
9
ACT Research & Policy | Technical Brief | May 2020
Table 4. Summary of the Relationship with First-Term GPA by Predictor
ACT Composite score
High school GPAStatistic Full test – most recent ACT Superscore*
n 118 118 114
Mean (SD) 30.3 (3.9) 31.0 (3.6) 3.82 (0.21)
Relationship with rst-term GPA
Regression estimates
Intercept** 3.36 (3.27, 3.45) 3.36 (3.27, 3.45) 3.35 (3.26, 3.44)
Slope** 0.18 (0.09, 0.27) 0.16 (0.07, 0.25) 0.15 (0.06, 0.24)
Correlation** .34 (.17, .49) .32 (.14, .47) .29 (.11, .45)
Note. GPA = grade point average. SD = standard deviation. For the regression estimates, the predictors were
standardized to have a mean of 0 and a standard deviation of 1. * ACT Superscore computed by combining the highest
subject scores across test administrations. ** The values reported include the estimate and the 95% condence interval
of the estimate in parentheses.
As shown in Figure 2 and Table 4, ACT Superscores were found to be predictive of rst-term
GPA. The regression lines and parameter estimates for ACT Superscores were similar to those
obtained from using students’ most recent ACT Composite score. This is evident by the 95%
condence intervals for the slopes associated with the standardized scores overlapping for
the two scoring methods. Additionally, there was not a signicant dierence in the correlation
coecients between the two scoring methods (Table 4; R = .34 for most recent and .32 for
Superscore; p-value = .205). These correlations are consistent with that computed for the
institution’s entire freshman cohort based on students’ most recent ACT Composite score
earned while in high school (R = .30, 95% condence interval = .26, .33).
Figure 2. Predicted First-Term GPA by ACT Composite Score and Scoring Method
2.00
2.20
2.40
2.60
2.80
3.00
3.20
3.40
3.60
3.80
4.00
21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT Composite score
Full test - most recent Superscore
10
ACT Research & Policy | Technical Brief | May 2020
To further examine the predictive validity of using ACT Superscores, we developed regression
models that included both ACT scores and high school GPA to estimate students’ rst-term
GPA. According to the single-predictor model shown in Table 4, high school GPA was correlated
with rst-term GPA for the participant sample (R = .29; 95% condence interval = .11, .45).
When including both ACT scores and high school GPA in the model (results shown in Table 5),
ACT Superscore continued to contribute to the prediction of rst-term GPA above and beyond
high school GPA. Specically, ACT Superscore was found to be signicantly related to rst-
term GPA in the joint model that included high school GPA (standardized coecient = 0.14;
95% condence interval = 0.05, 0.23). Additionally, ACT Superscore improved the prediction;
there was an increase in the validity coecient associated with the joint model as compared to
the model that only included high school GPA (R = .40 vs. .29, respectively). Moreover, when
results for ACT Superscore were compared to those obtained when students’ most recent ACT
Composite scores were used instead, we found the parameter estimates, multiple correlation
coecient, and increase in multiple correlation over high school GPA alone to be similar for the
two ACT Composite scoring methods (Table 5). The two multiple correlations reported in Table
5 for the study sample are consistent with that computed for the institution’s entire freshman
cohort using students’ most recent ACT Composite score from high school in combination with
high school GPA (R = .42, 95% condence interval = .39, .45).
Table 5. Summary of Joint Prediction Models for First-Term GPA by ACT Composite Scoring
Method
ACT Composite score
Statistic Full test – most recent ACT Superscore*
Regression estimates
Intercept** 3.35 (3.27, 3.44) 3.35 (3.27, 3.44)
ACT score slope** 0.16 (0.07, 0.25) 0.14 (0.05, 0.23)
High school GPA slope** 0.12 (0.03, 0.21) 0.13 (0.04, 0.22)
Multiple Correlation** .42 (.26, .56) .40 (.23, .54)
Increase in correlation over high
school GPA alone 0.13 0.11
Note. Model based on 114 participating students with a high school GPA available. For the regression estimates, the
predictors were standardized to have a mean of 0 and a standard deviation of 1. * ACT Superscore computed by
combining the highest subject scores across test administrations. ** The values reported include the estimate and the
95% condence interval of the estimate in parentheses.
11
ACT Research & Policy | Technical Brief | May 2020
Conclusions
In conclusion, this study provides additional support for the use of section test scores and ACT
Superscores for predicting students’ rst-year college grades. More specically, ndings from this
study indicate that ACT scores earned from section retests are as predictive of rst-term grades
as scores earned from taking the full test. Additionally, results from the study suggest that ACT
Superscores obtained from combining scores across test administrations including section retests
are as predictive of rst-term college grades as students’ most recent full test scores.
Having said that, we acknowledge that this study has limitations. First, this study was conducted in
collaboration with only one postsecondary institution. Second, the number of participating students
was relatively small overall and by subject area. This resulted in wide condence intervals for the
correlation coecients and the standardized ACT slope estimates. But, the regression results
for the sample were found to be similar to those based on the institution’s entire freshman
cohort of ACT-tested students, suggesting that we would have found similar results if a larger
number of students had participated from the institution. Moreover, the results from this study are
consistent with those from an earlier multi-institutional study by Mattern et al. (2018) that found
ACT Superscores obtained by combining the highest section scores across test administrations to
be as predictive of rst-year grades as other ACT Composite scoring methods. The earlier study
involved nearly 278,000 students attending 221 four-year institutions.
Because the section retest occurred during the rst three weeks of the fall term and was a
low-stakes testing event, it was somewhat expected that some students may not have been as
motivated or prepared for the section retest as they were when they took the full ACT test to
receive a college-reportable score and used it to gain admissions to colleges. To try to increase
participant motivation, we oered an extra monetary incentive to students who met or exceeded
their prior scores from high school. Additionally, analyses were conducted not only on the full
sample of participants but also on a subsample of students who did not experience a large
score decline on their section retest as compared to their latest full ACT score from high school.
The nding that section retest scores are as predictive of rst-term grades as scores obtained
via traditional full ACT testing was not dependent on the sample used in the analyses (full
sample vs. subsample). Unfortunately, we were unable to quantify and control for students’ level
of motivation in this study which could have impacted some of the results. Additionally, the small
sample size prevented examination of dierential validity by test administration (section vs. full
test) for demographic subgroups. Future studies will examine this topic.
Despite these limitations, the results based on this study suggest that section retest scores
and ACT Superscores that combine scores across test administrations including section retest
events are as predictive of rst-term college grades as students’ most recent full test scores.
Once the section retesting and superscoring options become operational in September 2020,
ACT will work with interested institutions to reexamine these issues. ACT is committed to
continuing its eorts in conducting national validity studies to provide evidence supporting the
use of ACT section retest scores and ACT Superscores, in combination with other measures, for
college admission and course placement decisions and for identication of students who may
benet from additional academic services and supports once they matriculate to college.
12
ACT Research & Policy | Technical Brief | May 2020
Notes
1. ACT Superscores are typically higher than an ACT Composite score earned from any single test
attempt (Cruce & Mattern, 2020).
2. Initially, study eligibility required that a student be enrolled in a specic subject-relevant course
to take the corresponding section test (e.g., English Composition I to take the ACT English
section test). The reason for this requirement was so that we could examine section test scores
in relation to the grades earned in typical rst-year subject-relevant courses. However, this
eligibility requirement was found to be too restrictive as too few students from the selected
classes were interested in participating in the study. During the rst week of classes, the decision
was made to open participation in the study to any rst-time entering student who had previously
taken the ACT test in high school during the 2017–18 or 2018–19 academic year.
3. Two-thirds of the participants (or 67.0%) took only one single-subject section test; 33.0% took
multiple section tests (22.9% in two subjects, 5.9% in three subjects, and 4.2% in all four subject
areas).
4. For the math, reading, and science samples, there was one major outlier that scored 12 to 19
points lower on their section retest as compared to their latest score from high school. The
corresponding student was omitted from the subject-specic analyses as the inclusion of their
scores considerably elevated the correlation between ACT section scores and rst-term GPA,
especially in math. If the outlier was included in the subject sample, the correlations for the most
recent scores and section retest scores were .47 and .62 in math, .27 and .38 in reading, and
.30 and .38 in science, respectively.
5. The percentage of students scoring the same or higher on the section retest as compared to
their ACT test score from high school was 37% (English), 48% (math), 38% (reading), and 53%
(science).
6. For every subject except math, the correlations between the section retest scores and the full
test-most recent test scores for the subsample were estimated to be higher than those obtained
in an earlier study (Radunzel & Mattern, 2020; Table 2). We also found this to be the case when
the correlations between the section retest scores and the full test-most recent test scores were
compared to the test-retest Pearson correlation coecients for the institution’s entire freshman
cohort who took the full ACT more than once in high school (54.2% of the institution’s ACT-tested
sample). For the test-retest correlations, a student’s most recent score from high school was
compared to their closest prior ACT test score; the typical time between these two testing events
was 3.7 months. The test-retest correlation (and 95% condence interval) was .74 (.72, .76) in
English, .80 (.78, .82) in math, .67 (.64, .70) in reading, and .62 (.59, .65) in science. In English,
reading, and science, the correlations between the section retest scores and full test-most recent
scores were more similar to the high school test-retest correlations for the full sample than for
the subsample. In math, the correlation between the section retest scores and full test-most
recent scores was more similar to the high school test-retest correlation for the subsample than
for the full sample.
7. In math and science, the slopes for the section test scores could be slightly higher than those for
the full test scores from high school due to the scores being obtained more proximal to when the
college courses were taken and earned. But, this could also be an artifact of the small sample
size.
8. The corresponding percentages by subject area were 3.4% in English, 9.3% in math, 5.1% in
reading, and 11.0% in science. These percentages do not sum to the total percentage because
students could have increased their scores on the section retest in multiple subject areas.
13
ACT Research & Policy | Technical Brief | May 2020
References
ACT. (2019). ACT technical manual. Iowa City, IA: ACT.
Andrews, B. (2019). Initial evidence in support of section retakes: The impact of administering
the ACT subject tests in dierent orders on ACT scores. Iowa City, IA: ACT.
Clinedinst, M. (2019). 2019 state of college admission. Alexandria, VA: National Association for
College Admissions Counseling.
Cruce, T., & Mattern, K. (2020). The impact of superscoring on the distribution of ACT scores.
Iowa City, IA: ACT.
Harmston, M., & Crouse, J. (2016). Multiple testers: What do we know about them? Iowa City,
IA: ACT.
Mattern, K., & Allen, J. (2016). More information, more informed decisions: Why test-optional
policies do NOT benet institutions or students. Iowa City, IA: ACT.
Mattern, K., & Radunzel, J. (2019). Does superscoring increase subgroup dierences? Iowa
City, IA: ACT.
Mattern, K., Radunzel, J., & Andrews, B. (2019). An initial look: Taking ACT subject tests on
dierent days doesn’t result in higher than expected scores. Iowa City, IA: ACT.
Mattern, K., Radunzel, J., Bertling, M., & Ho, A. D. (2018). How should colleges treat multiple
admissions test scores? Educational Measurement: Issues and Practice, 37(3), 11−23.
Radunzel, J. (2017). Using incoming student information to identify students at-risk of not
returning to their initial institution in year two. Iowa City, IA: ACT.
Radunzel, J., & Mattern, K. (2020). Section retesting: Do students perform as expected? Iowa
City, IA: ACT.
Sanchez, E., & Buddin, R. (2016). How accurate are self-reported high school courses, course
grades, and grade point average? Iowa City, IA: ACT.
Sawyer, R. (2010). Usefulness of high school average and ACT scores in making college
admission decisions. Iowa City, IA: ACT.
Steiger, J. H. (1980). Tests for comparing elements of a correlation matrix. Psychological
Bulletin, 87(2), 245-251.
Westrick, P. A., Le, H., Robbins, S. B., Radunzel, J. M. R., & Schmidt, F. L. (2015). College
performance and retention: A meta-analysis of the predictive validities of ACT
®
scores, high
school grades, and SES. Educational Assessment, 20(1), 23−45.
University of California Academic Senate. (2020). Report of the UC Academic
Council Standardized Testing Task Force (STTF). Retrieved from https://senate.
universityofcalifornia.edu/_les/underreview/sttf-report.pdf on February 26, 2020.
U.S. Department of Education, National Center for Education Statistics. (2018-2019). Integrated
Postsecondary Education Data System (IPEDS). Retrieved from https://nces.ed.gov/ipeds/
datacenter/InstitutionByName.aspx on February 19, 2020.
14
ACT Research & Policy | Technical Brief | May 2020
Appendix
Table A1. Percentages of Students by Attribute and Sample
Variable
English
sample
Math
sample
Reading
sample
Science
sample
ACT-tested
nonparticipants from
institution
Sample size 39 39 50 46 2,729
Gender
Female 56.4
23.1
44.0 47.8 51.8
Male 43.6
76.9
56.0 52.2 48.2
Race/ethnicity
Underserved minority 30.8
15.4
10.0 19.6 17.9
White 48.7
46.2
64.0 50.0 59.1
Asian 12.8
28.2
14.0 17.4 13.5
Multiple/unknown race 7.7
10.3
12.0 13.0 9.5
Annual family income
Less than $36,000 12.8
12.8
4.0 8.7 4.7
$36,000 to $80,000 18.0
15.4
14.0 19.6 11.1
More than $80,000 43.6
38.5
42.0 37.0 49.4
Missing 25.6
33.3
40.0 34.8 34.8
Educational aspirations
Below bachelor’s degree 0.0 0.0 0.0 0.0 0.3
Bachelor’s degree 35.9
28.2
34.0 23.9 34.2
Beyond bachelor’s degree 56.4
59.0
54.0 65.2 54.5
Missing 7.7
12.8
12.0 10.9 11.0
Declared major category
Non-STEM 33.3 12.8 30.0 15.2 38.3
STEM 59.0 79.5 58.0 67.4 54.1
Undecided/unknown 7.7 7.7 12.0 17.4 7.7
Returned for second term
Yes 100.0 97.4 100.0 100.0 99.4
No 0.0 2.6 0.0 0.0 0.6
Note. Student characteristics provided by institution or obtained from the ACT student record. Underserved minority
students included African American, American Indian, Hispanic, and Native Hawaiian or other Pacic Islander. Science,
technology, engineering, and mathematics (STEM) majors included science majors, computer science & mathematics
majors, engineering & technology majors, and medical & health majors.
15
ACT Research & Policy | Technical Brief | May 2020
Table A2. Linear Regression Parameter Estimates for Predicting First-Term GPA by Subject,
Sample, and Testing Event
Subject Predictor
Full test – most recent Section test
Est. SE
95%
Lower
95%
Upper Est. SE
95%
Lower
95%
Upper
English
Full sample Intercept 3.37 0.07 3.24 3.51 3.37 0.06 3.24 3.50
ACT score 0.15 0.07 0.01 0.28 0.18 0.06 0.05 0.31
Subsample Intercept 3.39 0.08 3.23 3.55
3.39 0.08 3.23 3.56
ACT score 0.20 0.08 0.03 0.37
0.20 0.08 0.03 0.37
Math
Full sample Intercept 3.39 0.08 3.24 3.54 3.39 0.07 3.24 3.53
ACT score 0.23 0.08 0.08 0.39 0.28 0.07 0.13 0.43
Subsample Intercept 3.40 0.09 3.22 3.58 3.40 0.08 3.23 3.57
ACT score 0.22 0.09 0.04 0.40 0.28 0.08 0.10 0.45
Reading
Full sample Intercept 3.46 0.06 3.34 3.59 3.46 0.06 3.34 3.59
ACT score 0.11 0.06 -0.02 0.24 0.14 0.06 0.02 0.27
Subsample Intercept 3.51 0.07 3.38 3.64 3.51 0.07 3.38 3.64
ACT score 0.12 0.07 -0.01 0.26 0.13
0.07 -0.01 0.26
Science
Full sample Intercept 3.41 0.07 3.26 3.56 3.42 0.07 3.27 3.56
ACT score 0.16 0.07 0.01 0.31 0.19 0.07 0.04 0.34
Subsample Intercept 3.40 0.09 3.23 3.58 3.41 0.08 3.24 3.58
ACT score 0.16 0.09 -0.02 0.34 0.21 0.08 0.04 0.39
Note. GPA = grade point average. Est. = estimate. SE = standard error. 95% Lower and 95% Upper corresponds to
the lower and upper limits for the 95% condence intervals. ACT subject scores were standardized to have a mean of
0 and a standard deviation of 1. The full sample includes all students’ scores except the one outlier (see endnote #4).
The subsample excludes students whose scores on section retesting decreased by more than 2 SEM compared to their
latest full test score from high school. Students completed the entire ACT test during their junior or senior year in high
school, while students completed the section retest during the rst three weeks of their freshman year in college.
16
ACT Research & Policy | Technical Brief | May 2020
Figure A1. Predicted First-Term GPA by ACT Section Score and Testing Event for Full Samples
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT English score
Full test - most recent Section test
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT math score
Full test - most recent Section test
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT reading score
Full test - most recent Section test
2.00
2.40
2.80
3.20
3.60
4.00
15 17 19 21 23 25 27 29 31 33 35
Predicted first-term GPA
ACT science score
Full test - most recent Section test
Acknowledgements
The authors thank Je Allen and Wayne Camara for their input and suggestions on earlier versions
of this brief.
Justine Radunzel, PhD
Justine Radunzel is a principal research scientist in Validity and Ecacy Research specializing in
postsecondary outcomes research and validity evidence for the ACT test.
Krista Mattern, PhD
Krista Mattern is a senior director in Validity and Ecacy Research whose research focuses on
predicting education and workplace success through evaluating the validity and fairness of cognitive
and non-cognitive measures. Krista is also known for her work in evaluating the ecacy of learning
products to help improve intended learner outcomes.
