Sunteți pe pagina 1din 19

Res High Educ (2008) 49:6279

DOI 10.1007/s11162-007-9064-5

Prediction of College Major Persistence Based on


Vocational Interests, Academic Preparation, and First-
Year Academic Performance

Jeff Allen Steven B. Robbins

Received: 27 October 2006 / Published online: 4 August 2007


! Springer Science+Business Media, LLC 2007

Abstract We hypothesized that college major persistence would be predicted by first-


year academic performance and an interest-major composite score that is derived from a
students entering major and two work task scores. Using a large data set representing 25
four-year institutions and nearly 50,000 students, we randomly split the sample into an
estimation sample and a validation sample. Using the estimation sample, we found major-
specific coefficients corresponding to the two work task scores that optimized the pre-
diction of major persistence. Then, we applied the estimated coefficients to the validation
sample to form an interest-major composite score representing the likelihood of persisting
in entering major. Using the validation sample, we then tested a theoretical model for
major persistence that incorporated academic preparation, the interest-major composite
score, and first-year academic performance. The results suggest that (1) interest-major fit
and first-year academic performance work to independently predict whether a student will
stay in their entering major and (2) the relative importance of two work task scores in
predicting major persistence depends on the entering major. The results support Hollands
theory of person-environment fit and suggest that academic performance and interest-major
fit are key constructs for understanding major persistence behavior.

Keywords College major persistence ! Hollands theory ! Person-environment fit !


College GPA ! Hierarchical logistic regression

Introduction

The choice of major is one of the most important decisions students make as they enter
college. Recent theories suggest that students who choose a major congruent with their
skills and interests are more likely to succeed and persist in that major and in college.

J. Allen (&) ! Steven B. Robbins


ACT Research, ACT, Inc, Iowa City, IA 52243-0168, USA
e-mail: jeff.allen@act.org

123
Res High Educ (2008) 49:6279 63

Smart, Feldman, and Etherington (2000, p. 33) suggest that students choose academic
environments compatible with their personality types. Further, they suggest that aca-
demic environments reward different patterns of student abilities and interests. These
suggestions led us to two propositions: (1) major choice is affected by a students pattern of
interests and (2) the students level of satisfaction and success with their academic envi-
ronment (major) will be affected by how well their interests fit the environment. Porter and
Umbach (2006) studied how several student characteristics, including interests, relate to
major choice. Their findings clearly supported the first proposition, as have other works
(Hackett and Lent 1992; Walsh and Holland 1992). As part of this study, we address the
second proposition by investigating how interest-major fit relates to third-year major
persistence.
We consider major persistence an important criterion as we believe it is an indicator of
satisfaction with academic environment: we believe that students who choose majors
consonant with their interests are more likely to persist in their chosen major. A students
satisfaction with their academic environment is important for the sake of their academic
integration and commitment to the college (see Tinto 1993, pp. 112118 for discussion). In
addition to being an indicator of satisfaction with major choice, persistence in major is an
important criterion variable since it affects timeliness of graduation. Generally, students
who change majors are more likely to take courses not necessary for graduation. Only 51
percent of incoming students at four-year institutions are able to complete a bachelors
degree within six years (ACT 2001). While this rate varies by institution and degree area,
the aggregate rate is shockingly low and might be improved if more students did not
change major. While its certainly good for students to change majors in order to find an
environment more congruent with their interests and skills, we believe its even better for
students to begin their studies in their optimal environment. By studying how interests
predict major persistence, we hope to better guide students to their optimal environment
before they start college.

Interest patterns, major environment, and relations to major persistence

One component of Hollands theory of vocational preferences (1997) is that environments


are characterized by activity types in the same way that individuals are characterized by
personality types. Further, person-environment fit can be assessed by the congruence of an
individuals personality type to their environments activity types. This theory gives rise to
Smart et al. (2000) proposal that academic environments reward different patterns of
student abilities and interests. In terms of major persistence, this suggests that a students
interest-major fit will impact their likelihood of persisting in that major.
The literature on person-environment fit and its relation to various criteria is abundant,
but few studies relate directly to college major persistence. Kramer et al. (1994) tracked
multiple cohorts at a four-year institution and found that nearly 75% of students changed
their major. They suggest that many students moved to majors more compatible with their
skills and interests. Slaney (1984) demonstrated that college women who had stable college
majors were more likely to report higher GPAs and stronger relations between expressed
and measured interests. Porter and Umbach (2006) studied the relations of four of the six
Holland types (investigative, artistic, social, and enterprising) to college major choice.
They found that race/ethnicity, major uncertainty, political views, and the four personality
types were related to selection of major type. Generally, students higher in investigative,
lower in artistic, and lower in enterprising were more likely to choose a science major.

123
64 Res High Educ (2008) 49:6279

Although they did not study major persistence or satisfaction with entering college
major, Lounsbury et al. (2005) did study the relation of personality traits and overall level
of college satisfaction. They found that work drive, emotional stability, career decidedness,
aggression, and optimism were jointly predictive of college satisfaction. This work has
important implications for our study since it shows that factors other than first-year aca-
demic performance and student-environment fit affect college satisfaction (and hence may
also affect major persistence). Unfortunately, our data set does not allow us to directly
measure these same constructs and so we did not include them as predictors in our model
for major persistence. Pike (2006) analyzed the relationships between students personality
types, expectations about college, and intended majors. His findings suggest that students
expectations of college activities influence the selection process of entering major. His
work did not address major persistence, but supports the idea that students expectations
and personality types should fit their academic environment.
Other literature has involved person-environment fit and its relation to other college
outcomes. Tracey and Robbins (2006) examined the interest-major congruence and college
success relationship using longitudinal data from a large sample of students and postsec-
ondary institutions. After adjusting for academic preparation, they found that greater
interest-major congruence was associated with higher rates of retention. Tracey and
Robbins also found that interest-major congruence was predictive of graduation status.
They did not examine major stability, nor did they build models that incorporated first-year
college GPA when predicting outcomes beyond the first year.
Career development literature suggests that academic skills alone are not enough to
promote career satisfaction (cf. Dawis and Lofquist 1984), but that person-environment fit
must also be considered. For example, adult workers who re-entered postsecondary edu-
cation and chose college majors and career fields congruent with their interests and abilities
were more likely to be satisfied (Kristof-Brown et al. 2005). Work by Olitsky et al. (2007)
suggests that congruence of interests and planned college major, when measured before
college, has a positive relationship with post-collegiate earnings. Donohue (2006) reported
that a group of career changers had lower congruence scores (using the C-Index developed
by Brown and Gore 1994) than a group of career persisters. Further, the career changers
moved towards careers that were more congruent with their personality profiles. Finally,
those with greater vocational-interest fit are more likely to persist within their career
(Oleski and Subich 1996).

The Roles of First-Year College GPA and Academic Preparation

Logically, students who perform well in their entering college major are more likely to
persist in that major. Leuwerke et al. (2004) found that academic preparation, as measured
through the ACT Mathematics score, predicted whether entering engineering students
persisted in their major. Their study did not assess the relations of high school GPA and
first-year GPA on major persistence. Though the relations of academic preparation and
major persistence are not well established, past work has shown strong relations of aca-
demic performance and college persistence. In their review of the college success
literature, Pascarella and Terenzini (2005) note the central role first-year college GPA
plays in predicting educational attainment and persistence. Allen et al. (2007) examined
the roles of first-year GPA and pre-collegiate measures of psychosocial factors in pre-
dicting retention to the third year. They found that first-year GPA was a strong predictor of
retention, and that academic-specific motivation and social engagement also had direct

123
Res High Educ (2008) 49:6279 65

effects on retention. Porter and Umbach (2006) found that SAT scores did not predict
major choice, but this sheds little light on the relation of high school academic preparation
and major persistence. High school academic preparation relates to college academic
performance and retention behavior, with standardized achievement test score and high
school GPA typically accounting for 28% of the variance of first-year college GPA (ACT
1997). Recent work reinforces the relations of high school GPA and achievement test score
on college GPA and persistence, and suggests that measured psychosocial factors also
contribute to the prediction (see Robbinset al. 2006; Allen et al. 2007).

Study Hypotheses

Our first hypothesis is that major persistence is predicted by interest-major fit. In the
Method section, we will discuss how we capture vocational interest patterns by using
scores on two dimensions of work tasks. Consistent with Hollands theory of person-
environment fit, we propose that the importance of each work task will vary across entering
major groups. For example, the vocational interest pattern lending itself to sustaining an
engineering major may be quite different than the interest pattern compatible with sus-
taining an English major.
Our second hypothesis is that students with higher first-year GPA are more likely to
persist in their entering major. Given the past work relating first-year GPA to college
persistence and our belief that academic performance and satisfaction with major feed one
another, this hypothesis seems logical.
Our third hypothesis is that indicators of pre-collegiate academic preparation (high
school GPA and ACT Composite score) are related to major persistence, but are mediated
by first-year GPA. Since high school academic preparation predicts first-year college GPA,
we expect that high school academic preparation will only have an indirect relationship
with major persistence.
The theoretical model that we hypothesize and test is summarized in Fig. 1. The
hypothesized relationships are given by the pointed lines: we expect interest-major fit and
academic performance to predict major persistence (hypotheses 1 and 2), and we expect
pre-collegiate academic performance to predict academic performance but not major
persistence (hypothesis 3).

Method

Sample Data

Our initial sample included 87,993 first-time entering students at 25 four-year institutions.
The outcome of interest was third-year college major persistence. In order to ascertain a
students third-year major, the student must have been enrolled during the third year;
hence, students who dropped out before or during the third year were excluded from the
analysis. Further, a students entering and third-year major must be known in order to
ascertain major persistence. Therefore, students who did not declare an entering major or
who were undecided were excluded from the analysis. Of our initial sample, 23,062
(26.2%) were not enrolled during the third year and so major persistence could not be
ascertained. Of the remaining 64,931 students, 16,699 (25.7%) did not declare an entering
major or were undecided. This left us with 48,232 students. Since our analysis involved

123
66 Res High Educ (2008) 49:6279

Pre-collegiate
Academic Preparation Academic Performance

Interest-Major Fit Major Persistence

Fig. 1 Theoretical model for predicting major persistence

estimating coefficients that are specific to major groups, we wanted to have a reasonably
large sample size within each entering major group (later, we will discuss the categori-
zation of majors). We excluded entering major groups with fewer than 100 students,
leaving us with a final sample of 47,914 students with complete data.
We now compare the final and excluded samples with respect to demographics, pre-
collegiate academic achievement, and college outcomes (first-year GPA and major per-
sistence). The samples represent students who entered college between 1994 and 2002,
with more students coming from the later years. Reflecting national averages (Snyder et al.
2004), the samples had more women (55.6%) than men. The final sample was mostly
Caucasian (79.2%), with smaller representation from African-American (6.3%), Asian-
American (5.1%), and Hispanic students (3.4%). Minority students were more likely to be
excluded due to dropout or undecided major. Students with lower high school GPA, lower
ACT score, and lower first-year GPA were more likely to be excluded due to dropout or
undecided major: the mean high school GPA was 3.40 in the final sample and 3.16 in the
excluded sample, the mean ACT Composite score was 24.1 in the final sample and 22.3 in
the excluded sample, and the mean first-year college GPA was 2.91 in the final sample and
2.53 in the excluded sample. Students in the excluded sample were slightly more likely to
be unsure about their entering major choice: 24.3% of the excluded sample were not
sure as compared to 18.6% of the final sample. The students in both samples were mostly
traditional first-time freshman enrollees98.6% of the sample was age 19 or younger
when they entered college. As we will discuss later, the differences in the final and
excluded samples have implications for the generalizability of this study.
ACT, Inc. has maintained relationships with several postsecondary institutions for
purposes of studying pre-collegiate data (collected via the ACT Assessment Program)
and its relation with college outcomes. By virtue of these relationships, the 25 insti-
tutions provided each students identification number, first-year GPA, enrollment status
during the third year, major during the first year, and major during the third year.
Student records were linked to ACT archival data, which contains standardized
achievement test scores, high school grades, measures of vocational interests corre-
sponding to each Holland type, and demographic data. The institutions varied in size
and admissions policy (selectivity), with all coming from the Midwestern region of the

123
Res High Educ (2008) 49:6279 67

country. Institution size (number of full-time undergraduates) ranged from 672 to


28,947, with a mean of 8,810 and standard deviation of 6,505. Each institution was
classified according to its admissions policy. One institution was classified as open (all
high school graduates accepted, to limit of capacity), three were classified as liberal
(some freshmen from lower half of high school graduating class), eleven were classified
as traditional (majority of accepted freshmen in top half of high school graduating
class), nine were classified as selective (majority of accepted freshmen in top quarter of
high school graduating class), and one was classified as highly selective (majority of
accepted freshmen in top 10% of high school graduating class).

Categorization of College Majors

For this study, we sought a categorization of majors that (1) is specific enough so that
majors in each group have similar or overlapping curricula and (2) is general enough so
that each category has a reasonable sample size. To categorize majors, we used the
Classification of Instructional Programs (CIP; NCES 2002) families of majors, given by
the first two digits of the CIP code. In four cases, we collapsed two similar families into a
single family. These include Agriculture, Agriculture Operations, and Related Sciences
and Agricultural Sciences; Communication, Journalism, and Related Programs & Com-
munications Technologies/Technicians and Support Services; Engineering and
Engineering Technologies/Technicians; and Philosophy and Religious Studies and The-
ology and Religious Vocations. Table 1 lists each major group, as well as the sample size
and percentage of students who persisted in their entering major group. We used 23 groups
of college majors and they varied with respect to major persistence rates. Major groups
with low persistence include transportation and materials moving (43.1%) and mathe-
matics and statistics (46.9%). Groups with high persistence include security and protective
services (77.6%) and engineering (76.5%).

Defining Major Persistence

Our criterion of interest was third-year major persistence. As described earlier, students
who dropped out before or during the third year were excluded from the analysis.
Further, students with undeclared entering majors were excluded. By excluding these
students, our final sample is not representative of all first-year students; generally, the
final sample is better-prepared academically and hence more likely to persist in college.
Our major persistence outcome is dichotomousstudents either remained in their
entering major group into their third year or switched major groups at some time before
or during the third year. By defining major persistence in this way, we consider the
criterion as an indicator of satisfaction with entering major and also an indicator of
continuity of curricula (students that change major might remain in the same major
group with overlapping curricula). For example, a student who begins as a Veterinary
Medicine major (CIP = 51.24) but switches to Dentistry (CIP = 51.04) is classified as
having persisted in major group since their entering group (Health Professions and
Related Clinical Sciences) did not change. However, if the same student had switched
to Accounting (CIP = 52.03) instead of Dentistry, they would be classified as having not
persisted.

123
68 Res High Educ (2008) 49:6279

Table 1 Sample size, persistence rates, and re-estimated coefficients (Standard Errors) by major group
Major group (CIP) N % a^0 (intercept) a^1 (D/I) a^2 (T/P)

Agriculture (01 & 02) 1,512 74.9 "0.625 (0.336) 0.178 (0.062) 0.487 (0.065)
Natural resources & conservation (03) 251 53.8 "1.612 (0.382) "0.201 (0.156) 0.107 (0.139)
Architecture & related services (04) 449 68.8 "0.990 (0.348) "0.284 (0.126) 0.198 (0.122)
Communication (09 & 10) 3,007 70.3 "0.822 (0.334) 0.155 (0.045) "0.016 (0.046)
Computer & inform. sci. & supp. serv. 2,137 57.9 "1.542 (0.334) "0.079 (0.052) 0.268 (0.053)
(11)
Education (13) 6,296 72.3 "0.779 (0.332) 0.117 (0.032) "0.056 (0.032)
Engineering (14 & 15) 5,057 76.5 "0.704 (0.332) 0.094 (0.040) 0.232 (0.039)
Foreign lang., literatures, & Linguistics 320 61.3 "1.316 (0.358) 0.314 (0.131) 0.084 (0.122)
(16)
Family & consumer sci./human sci. (19) 764 64.1 "1.010 (0.341) 0.090 (0.087) 0.116 (0.088)
English language & literature/letters (23) 1,406 66.6 "1.129 (0.338) "0.075 (0.058) "0.021 (0.064)
Biological & biomedical sciences (26) 3,805 63.6 "1.388 (0.333) "0.126 (0.039) 0.172 (0.039)
Mathematics & statistics (27) 746 46.9 "1.944 (0.341) "0.110 (0.08) "0.022 (0.079)
Parks, rec., leisure, & fitness studies (31) 635 76.2 "0.535 (0.343) "0.042 (0.107) "0.024 (0.110)
Physical sciences (40) 952 54.9 "1.712 (0.340) "0.065 (0.075) 0.108 (0.074)
Psychology (42) 2,191 65.9 "1.033 (0.334) "0.027 (0.050) 0.074 (0.053)
Security & protective services (43) 1,326 77.6 "0.298 (0.338) 0.041 (0.079) 0.255 (0.074)
Public admin. & soc. serv. professions 228 57.9 "1.452 (0.368) 0.101 (0.163) "0.166 (0.158)
(44)
Social sciences (45) 2,170 56.2 "1.469 (0.333) 0.124 (0.046) 0.027 (0.047)
Transportation & materials moving (49) 204 43.1 "1.965 (0.365) "0.209 (0.190) 0.167 (0.163)
Visual & performing arts (50) 2,728 72.2 "0.87 (0.335) "0.165 (0.052) 0.066 (0.047)
Health prof. & related clinical sci. (51) 2,285 67.6 "1.13 (0.333) "0.174 (0.055) "0.001 (0.052)
Business, mgmt., & marketing (52) 9,098 75.1 "0.701 (0.331) 0.256 (0.027) 0.160 (0.028)
History (54) 347 73.5 "0.782 (0.354) "0.022 (0.139) "0.038 (0.139)

Summarizing Vocational Interests in Two Dimensions

ACTs UNIACT interest inventory is offered to students when they register for the ACT.
This battery of vocational interests results in reliable (ACT 1995) scale scores for six basic
types of vocational interests, each of which corresponds to one Holland (1997) interest
type. The six scales (and corresponding Holland type) are: Science and Technology
(Investigative), Arts (Artistic), Social Service (Social), Business Administration and Sales
(Enterprising), Business Operations (Conventional), and Technical (Realistic). Each of
these six interest scales consists of 15 items of work-related activities. The response
options for each item are dislike, indifferent, and like. Each scale score, ranging
from 20 to 80, represents the level of interest for each vocational type.
Prediger (1982) suggested that two bipolar work task dimensions underlie Hollands
hexagon of interest types (Holland 1997)the Data/Ideas dimension and the Things/
People dimension. Other works (Tokar and Swenson 1995; ACT 1995) have supported this
idea and the two dimensions are used as the basis of ACTs World-of-Work Map (ACT
2001; Prediger and Swaney 2004). Tracey and Robbins (2005, 2006) argued for use of the

123
Res High Educ (2008) 49:6279 69

ADMINISTRATION & SALES BUSINESS OPERATIONS


(Enterprising) (Conventional)

Data
SOCIAL SERVICE People Things TECHNICAL
(Social) (Realistic)

ARTS Ideas SCIENCE & TECHNOLOGY


(Artistic) (Investigative)

Fig. 2 Relationship between Interest Scales (Holland types) and work task dimensions

World-of-Work Map and its underlying Data/Idea and People/Things coordinates as a


means of measuring the distance between an individuals measured interests and their
choice of either a college major or career area. They found that a Euclidean distance
measure was consistently related to college and high school outcomes. Olitsky et al. (2007)
also found the Euclidean distance measure to be predictive of post-collegiate earnings.
There are other studies that have used the World-of-Work Map and its underlying Data/
Ideas and People/Things dimensions as a basis for measuring congruence of vocational
interests and college major (cf. Prediger and Vansickle 1992; Rounds and Tracey 1993).
Figure 2 displays the relationship between the six scales of vocational interests (Holland
types) and the two work task dimensions. The six scales of vocational interests can be used
to obtain Data/Ideas and Things/People scores (ACT 1995), using the following equations:
Equation 1: Work Task Scores

Data/Ideas score "1:73I " 1:73A 1:73E 1:73C


Things/People score 2:00R 1:00I " 1:00A " 2:00S " 1:00E 1:00C

In Eq. 1, R represent the students Technical (Realistic) score, I = the Science and
Technology (Investigative) score, A = Arts (Artistic), S = Social Service (Social),
E = Business Administration and Sales (Enterprising), and C = Business Operations
(Conventional). With these scores, a students vocational interests can be characterized by
two work task scores that are consistent with the two dimensions underlying Hollands
hexagon.
In the total sample, the mean D/I score was "3.27 (standard deviation 36.26) and the
mean T/P score was "2.03 (standard deviation 33.94). We estimated internal reliabilities
of .88 for Technical (Realistic), .91 for Science and Technology (Investigative), .87 for
Arts (Artistic), .85 for Social Service (Social), .86 for Business Administration and Sales
(Enterprising), .91 for Business Operations (Conventional), .89 for D/I scores, and .86 for
T/P scores.

123
70 Res High Educ (2008) 49:6279

Estimation of Major-Specific Coefficients for Two Work Task Scores

We now discuss the estimation of major-specific coefficients for the two work task scores
(Data/Ideas and Things/People). We thought it was prudent to summarize students
interests in two dimensions rather than considering all six of Hollands personality types.
By considering just two work task scores, we will have fewer coefficients to estimate for
each major group, resulting in a more parsimonious and stable model. Further, the work by
Prediger (1982) and others suggests that little information is lost when summarizing
Holland-type interests with the Data/Ideas and Things/People scores.
As discussed earlier, we hypothesized that vocational interests (summarized by the work
task scores) will predict major persistence differentially across majors. This notion is
consistent with Hollands (1997) theory of person-environment fit. In simple terms, the
interests that are conducive to persisting in any given major group will depend on that
majors activity tasks. We used approximately half of our final sample to estimate major-
specific coefficients for the two dimensions of work tasks. Later, we will explain how the
other half of the sample was used to validate the coefficients for prediction of major
persistence. Now, we will explain the modeling strategy for estimating the major-specific
coefficients for the two work task scores.
In our data set, students are nested within institutions and so we used a hierarchical
logistic regression model (see Raudenbush and Bryk 2002) to account for the institutional
sources of variation. In order to estimate the major-specific coefficients, we allowed the
effects of the work task scores (Data/Ideas and Things/People) to vary by major group.
This was done by using D/I and T/P-by-major group interaction terms. The model assumes
that the ith student (from the jth institution and the kth major group) has probability pijk of
staying in their entering major group where pijk = { 1 + exp("lijk ) }"1 and lijk represents
the log-odds of staying in entering major group, where

Equation 2: Log-Odds of Major Persistence for Estimation Model


X
23
lijk bj GPAi b SUREi h Ii2k a0K DIi a1K TPi a2K :
k1

In Eq. 2, bj represents the intercept for the jth institution, GPAi represents the first-year
GPA for the ith student with corresponding regression parameter b, SUREi represents the
sureness of major choice for the ith student with corresponding regression parameter h, Ii[k
is an indicator for whether the ith student belongs to the kth major group (Ii[k = 1 if the ith
student belongs to the kth major group, = 0 otherwise), a0K represents the intercept for the
kth major group, DIi represents the Data/Ideas score for the ith student, a1K represents the
regression coefficient for Data/Ideas for the kth major group, TPi represents the Things/
People score for the ith student, and a2K represents the regression coefficient for Things/
People for the kth major group. We standardized the Data/Ideas and Things/People scores
to have mean zero and standard deviation one. Later, we will see how this simplifies the
interpretation of the major-specific coefficients (a0K ,a1K ,a2K ). In an attempt to explain the
institutional variation, we modeled the institution-specific intercepts with the institutions
enrollment size (number of full-time undergraduates) and admissions policy (1 = open,
2 = liberal, 3 = traditional, 4 = selective, 5 = highly selective). Hence, bj = k0 + sizej
k1 + policyj k2 + ej where k0 is the overall intercept and ej is the residual effect of the
jth institution, assumed to be normally distributed with mean 0 and unknown variance.

123
Res High Educ (2008) 49:6279 71

Students were assigned to the estimation or validation sample with equal probability
using pseudo-random number generation. Using the estimation sample, the model was fit
using the SAS GLIMMIX macro which uses iteratively reweighted likelihoods to fit the
generalized linear mixed model (Wolfinger and OConnell 1993). By fitting this model, we
obtained estimates of the major-specific coefficients for the two work task scores. These
coefficients represent the relative importance of each work task score in predicting a
students persistence in each major group. Since we believe that first-year academic per-
formance and major sureness will also predict major persistence, we thought it was
important to include these as covariates in the estimation model to better calibrate the
estimation of the major-specific coefficients of the work task scores.

A Composite Predictor of Major Persistence Based on Interests and Major

The model we fit for the estimation sample leads us to the formation of a composite
measure that predicts major persistence based upon a students measured interests and
entering major. After fitting the model using the estimation sample, we obtain estimates for
the major-specific coefficients, denoted a^0K ; a^1K ; a^2K for the kth major group (there are
23 sets of the coefficients). Using these estimated coefficients, a composite measure of the
likelihood of persisting in entering major is calculated as

Equation 3: Interest-Major Composite for Predicting Major Persistence


X
23
Ci Ii2k a^0K DIi a^1K TPi a^2K :
k1

In this expression, a^0K captures the overall likelihood of persisting in the kth major
group, a^1K captures the degree to which the likelihood of persisting in the kth major group
is influenced by students level of interest in Data/Ideas tasks, and a^2K captures the degree
to which the likelihood of persisting in the kth major group is influenced by students level
of interest in Things/People tasks. To test the predictive validity of the composite measure,
we calculated Ci for the students in the validation sample and used it as a predictor of
major persistence. By summarizing the effects of interests and major with a single pre-
dictor variable, we are better able to assess the strength of the relationship of interest-major
fit and major persistence. Further, since we have collapsed the major-specific intercepts
and work task coefficients into a single composite measure, there are fewer parameters to
estimate in the validation model.

Modeling Major Persistence in the Validation Sample

Using the validation sample, we fit models for first-year GPA and major persistence that
address our studys hypotheses. We first modeled first-year GPA using high school GPA
and ACT Composite score as predictors. Since students are nested within institutions, we
used a two-level hierarchical linear model and used the same institution-level predictors of
the intercept as was used for the estimation samples model (institution size and admissions
policy). Then, we modeled major persistence using the measures of pre-collegiate
academic preparation, first-year GPA, major sureness, and the interest-major composite

123
72 Res High Educ (2008) 49:6279

(Eq. 3) as predictors. We used a two-level hierarchical logistic regression model with the
same institution-level predictors described earlier.
Hypothesis 1 is supported if the interest-major composite is predictive of major per-
sistence the larger the size of the regression coefficient, the more the hypothesis is
supported. Hypothesis 2 is assessed by the strength of first-year GPA in predicting major
persistence. Hypothesis 3 is supported if high school GPA and ACT Composite score
predict first-year GPA, first-year GPA predicts major persistence (Hypothesis 2), and high
school GPA and ACT Composite score have little strength in predicting major persistence
above that already predicted by first-year GPA.

Predictor Variables

High school GPA, ACT Composite score, and major sureness were obtained by matching
to ACTs archival data. High school GPA was calculated using a students self-reported
grades over all four years of school to derive a cumulative GPA on a 4-point scale. The
ACT Composite score, a measure of overall performance on the ACT Tests of Educational
Development (mathematics, science, English, and reading tests), is a reliable (typical
reliability of .96) and valid predictor of first-year college grades (ACT 1997). When
students register for the ACT, they are asked about their educational plans. One question
asks how sure are you about your current choice of college major? The possible
responses are (1) very sure, (2) fairly sure, and (3) not sure. Since students who are less
sure of their college major seem more likely to change majors, we included this variable as
a predictor of major persistence. This predictor variable was coded as 1 = not sure,
2 = fairly sure, and 3 = very sure.

Results

Estimation Model

We begin by discussing the modeling results of the estimation model (see Eq. 2), which
was used to estimate the coefficients (a0K, a1K, a2K ) for each major group. The estimated
coefficients, in turn, will be used to form the interest-major composite predictor of major
persistence.
Since the students were randomly assigned from the total sample to the estimation or
validation sample, the two samples are of nearly the same size23,894 students were
assigned to the estimation sample and 24,020 to the validation sample. The estimation
model yielded coefficients a^0K ; a^1K ; a^2K for each major group, as well as their standard
errors. The a^1 values represent the degree to which students interests in Data/Ideas leads
to persistence in each major group. For example, a^1 0:244 for business majors. The
positive value for this coefficient means that students with interests that are high in Data
(as opposed to Ideas) are more likely to persist in a business major. Since the Data/Ideas
and Things/People scores were standardized and a logistic regression model was used, we
can interpret a^1 0:244 as the odds of persisting in a business major increase by a factor
of 1.28 (exp.244) for each standard deviation increase in Data/Ideas score. As another
example, a^2 0:217 for engineering majors: students with interests that are high in
Things (as opposed to People) are more likely to persist in an engineering major. For
some major groups, the likelihood of persisting does not appear to be impacted much by

123
Res High Educ (2008) 49:6279 73

students interests. For example, a^1 :061 and a^2 "0:013 for the History major group.
Since these coefficients are near zero, this suggests that students D/I and T/P scores have
little impact on the likelihood of persisting in this major group. For other major groups, the
likelihood of persisting appears to be impacted by both work task scores. For example,
both work task scores are important for persisting in business (^a1 0:244, a^2 0:170) and
agriculture (^a1 0:269, a^2 0:494) majors. Naturally, the standard errors of the coeffi-
cients are larger for major groups with smaller sample sizes. For example, a^1 "0:149 for
Transportation and Materials Moving, but the standard error of this estimate is large
(0.324) due to the small sample size (n = 204 in the total sample). By comparison, the
standard error of a^1 for Education is much smaller (0.046) due to the larger sample size
(n = 6,296).
Using the coefficients derived from the estimation sample, we calculated the interest-
major composite score according to Eq. 3, for the validation sample. The mean of the
composite was "0.491 with standard deviation 0.440.

Validation Sample: Correlations

Before considering the tests of the theoretical model (Fig. 1), we now present the simple
correlations of the predictors of interest (high school GPA, ACT Composite score, major
sureness, and interest-major composite) with first-year GPA and major persistence. These
correlations are presented in Table 2. Since major persistence is a dichotomous variable,
we didnt expect large correlations between this outcome and the predictors. With a sample
size of over 24,000, correlations of magnitude .022 and greater are statistically significant
at the .001 level. As expected, high school GPA and ACT Composite score were correlated
with first-year college GPA (r = .512, .440, respectively). Further, both measures were
correlated with major persistence (r = .093, .073), but it remains to be seen whether these
relations are due to mediation by first-year GPA. Major sureness was slightly related to
major persistence (r = .030). First-year GPA and the interest-major composite had little
relation with one another (r = ".041), but both were clearly related to major persistence:
the correlations were r = .157 for GPA and r = .160 for the interest-major composite.

Testing the Theoretical Model: Modeling Results for the Validation Sample

In this section, we explore how well the interest-major composite predicted major per-
sistence when the other predictors entered the model (first-year GPA, pre-collegiate
academic preparation, major sureness, and institutional characteristics). We also address
the studys hypotheses by examining the results of the validation model. In Table 2, we
present the estimated regression coefficients for the validation model. The institution-level
predictors were standardized at the institution level and the student-level predictors were
standardized at the student level so that the regression coefficients represent changes in the
outcome with respect to one standard deviation changes in the predictor. There was var-
iation in first-year GPA across institutions (variance = 0.009, P = .029), but neither
institution size (estimated regression coefficient of "0.068, P = .459) nor admissions
policy ("0.076, p = .024) did little to explain the variation. As expected, the most salient
predictors of first-year GPA were pre-collegiate academic preparation indicators (high

123
74 Res High Educ (2008) 49:6279

Table 2 Validation sample: correlations and estimated regression coefficients for validation model
Predictor variable First-year GPA Major persistence

EST SE P EST SE P

Institution level
Intercept 3.039 0.032 <.001 0.959 0.066 <.001
Enrollment size "0.068 0.049 .459 "0.066 0.072 .370
Admissions policy "0.076 0.028 .024 0.005 0.073 .944
Institutional variance 0.009 0.005 .029 0.093 0.034 .003
Student level r r
High school GPA .512 0.270 0.006 <.001 .093 0.001 0.021 .971
ACT composite score .440 0.195 0.006 <.001 .073 0.039 0.020 .060
Interest-major composite ".041 .160 0.383 0.015 <.001
Major sureness -.062 .030 0.100 0.015 <.001
First-year GPA .157 0.360 0.022 <.001
Multiple R .589 .252

school GPA and ACT Composite score). The model for first-year GPA resulted in an
overall correlation (Multiple R) of .589.
There was also significant variation across institutions in major persistence rates (var-
iance = 0.093, P = .003). Again, neither institution size (estimated logistic regression
coefficient of "0.066, p = .370) nor admissions policy (0.005, P = .944) explained much
of this variation. The interest-major composite did predict major persistence (0.383,
P < .001). The logistic regression coefficients represent the log-odds of persisting in major
related to a standard deviation increase in the predictor. So, we estimate that the odds of
persisting in major group increase by a factor of 1.47 (exp0.383) for each standard deviation
increase in the interest-major composite. Since the interest-major composite predicts major
persistence, hypothesis 1 is supported.
First-year GPA also predicted major persistence (0.360, p < .001). We estimate that the
odds of persisting in major group increase by a factor of 1.43 for each standard deviation
increase in first-year GPA. The effect size of first-year GPA is slightly less than the size of
the interest-major composite. This finding supports hypothesis 2.
As expected, major sureness predicted major persistence (0.100, P < .001), though its
effect size was much smaller than that of the interest-major composite and first-year GPA.
Neither high school GPA nor ACT Composite score contributed to the prediction of major
persistence in the validation model. Earlier, we saw that these measures of pre-college
academic preparation were positively correlated with major persistence and first-year GPA
(Table 2). But, it appears that their effects on major persistence are mediated by first-year
GPA. These findings support hypothesis 3.
The overall correlation (Multiple R) for the major persistence model was .252. This
represents the correlation of the models fitted probabilities and the dichotomous outcome.
As discussed by Menard (2000), R2 (and R) indices for models for dichotomous data are
difficult to interpret and are dependent on the overall proportion of successes (i.e., the
overall proportion of students who sustained their initial major). So, this measure of model
fit should be interpreted with care.

123
Res High Educ (2008) 49:6279 75

Re-estimation of Major-Specific Coefficients for Work Task Scores Using Total


Sample

For purposes of developing and validating the interest-major composite measure, we split
our total sample into an estimation sample and a validation sample. The results of the
validation model suggest that the interest-major composite does indeed predict major
persistence and that hypothesis 1 is supported. Approximately half of the total sample was
used to estimate the major-specific coefficients used to define the interest-major composite.
If we were to use the total sample instead, the estimated coefficients would have smaller
standard errors (generally, the standard errors should shrink by a factor of H 2 if the sample
size is doubled). In turn, the interest-major composite scores will be somewhat more
reliable. Therefore, we re-estimated the major-specific coefficients by re-fitting the esti-
mation model (Eq. 2) using the total sample. The resulting estimates are given in Table 1.
Generally, the re-estimated coefficients agreed with the original estimates. The correlation
of the original and re-estimated intercepts a^0K , D/I coefficients a^1K , and T/P coeffi-
cients a^2K was .95, .73, and .83, respectively. For groups with large sample sizes (i.e.,
business, education, and engineering), the estimated coefficients have smaller standard
errors and so are less likely to change upon re-estimation.

Discussion

Support of Hollands Theory

Our findings support Hollands theory of vocational preferences and the proposition that
students are more likely to flourish in academic environments that fit their personality types
(Smart et al. 2000; Pike 2006). We now know that interests affect both choice of entering
major (Porter and Umbach 2006) and the likelihood of persisting in major. We developed
and validated a composite score that measures the likelihood of persisting in major. We
refer to the composite score as the interest-major composite because it incorporates both
entering major and scores on six basic interest scales that correspond to Hollands per-
sonality types. By categorizing majors according to CIP family and collapsing six
vocational interests (Holland personality types) into two work task scores, we propose a
succinct method of measuring students fit in academic environment.

Interest-Major Composite and First-Year GPA are Key Predictors of Major Persistence

The results of the validation model (Table 2) show that the interest-major composite and
first-year GPA predicted major persistence. Since these two key predictors had little
correlation with one another (r = ".041), their relations with major persistence were not
affected by the others inclusion in the model. The estimated regression coefficients were
similar for the two predictors (0.383 for the interest-major composite and 0.360 for first-
year GPA) and suggest that student-major fit and academic performance are key constructs
for understanding major persistence behavior.
Our findings highlight the importance of pre-collegiate academic preparation as a
precursor to college success. ACT Composite score and high school GPA account for
much of the variation in first-year GPA, which in turn predicts major persistence. We
observed positive correlations of ACT score and high school GPA with major persistence,

123
76 Res High Educ (2008) 49:6279

and we believe this relationship is fully mediated by first-year academic performance.


While first-year GPA was clearly predictive of major persistence, we believe that students
grades in entering major are even more relevant. Students at four-year institutions typically
take several general education courses during their first year, so first-year GPA is often a
measure of performance in general courses rather than performance in major. Unfortu-
nately, our data set did not include specific course grades and so we could not test the
hypothesis that major-specific grades are more relevant than first-year GPA for predicting
major persistence. Even with this limitation, the importance of overall academic perfor-
mance was apparent.
Other work suggests that academic performance and person-environment fit are part of a
larger set of constructs needed to understand major persistence behavior. Lounsbury et al.
(2005) studied the relation of personality traits and overall level of college satisfaction.
They found that work drive, emotional stability, career decidedness, aggression, and
optimism were related to college satisfaction. Since we view major persistence as an
indicator of satisfaction with academic environment, it leads us to believe that our findings
can be used to complement Lounsbury et al.s. Future research should attempt to integrate
the two sets of findings and test a theoretical model for major persistence that includes fit
with academic environment, academic performance in major, and the personality traits
identified by Lounsbury et al. By studying personality traits in conjunction with student-
major fit and academic performance, we might also better understand why some students
with seemingly poor interest-major fit and poor academic performance still persist and,
conversely, why some students with seemingly good fit and good academic performance
change majors.

Implications for Advising Students

The interest-major composite can be calculated for anyone with a set of vocational interest
scores (or related measures of Holland types) and with a planned entering major that
belongs to one of the major groups (CIP families) we studied. Since the interest-major
composite predicts persistence in major, it could be used by institutions to identify the
students most likely to change majors. This information could aid institutions and aca-
demic departments in their mission of helping students find the environment most
consonant with their interests, ultimately helping them succeed in college.

Study Limitations

Even with a large sample of nearly 50,000 students, many D/I and T/P coefficients
(Table 2) had large standard errors due to small within-major group sample sizes.
Therefore, the D/I and T/P coefficients are imprecise for some major-groups, decreasing
the reliability of the interest-major composite. Despite this problem, the interest-major
composite was the strongest predictor of major persistence in the validation sample. With
smaller standard errors of the D/I and T/P coefficients, we will have an even better
understanding of student-major fit and be better equipped to counsel students. Future
research should attempt to measure the D/I and T/P coefficients with greater precision.
By using hierarchical modeling, we accounted for institutional variation in GPA and in
major persistence. We considered two level-two predictors: institution enrollment and
admissions policy (selectivity). Neither was significantly related to the outcomes. Since we

123
Res High Educ (2008) 49:6279 77

expect Hollands theory of person-environment fit to hold true at all types of institutions,
we expect that the results would be similar if the study were replicated for a sample of two-
year institutions. In this case, the relevant criterion might be graduation in entering major
versus graduation in some other major. Future research could test whether interests predict
major persistence to the same degree across institutions types (i.e., do institutional char-
acteristics moderate the relations of interests and major persistence?).
Generalizability of this study is limited to students who enroll at four-year institutions,
declare an initial major, and persist through their third year of college. Approximately one
quarter of our initial sample was discarded because they dropped out before their third year.
Future research should attempt to better understand college persistence behavior with respect
to students fit with their academic environment. We expect that students whose interests do
not fit their entering major are more likely to be dissatisfied with their major, leading to lower
overall satisfaction with college, and ultimately dropout. When designing our study, we
contemplated including the college dropouts in our did not persist category. But, since
students leave college for other reasons not related to satisfaction with entering major
environment (see Tinto 1993, pp. 84135), we decided that the major persistence outcome
would be more meaningful if we restricted it to students who persisted in college. Our study
sample did not include students with an undecided initial major. Given the significant number
of students who enter college with an undecided major, this group warrants further exami-
nation. For example, are students with undecided majors more likely to have undifferentiated
interest profiles? Using the measurement methodology of this study, a student with an
undifferentiated interest profile would have D/I and T/P scores close to 0.
Our large sample allowed us to use a fine categorization of majors. Still, one could
argue that an even finer, or different, categorization of majors could yield different results.
Within each of our major groups, there exist many different and specialized fields of study
(see CIP; NCES 2002). Hollands theory of vocational preferences (1997) suggests that
environments are characterized by activity types in the same way that individuals are
characterized by personality types. Therefore, our findings are impacted by the within-
group homogeneity of major activities. For major groups that contain majors with very
similar activity types, we would expect the categorization to yield stronger results.
Finally, our study is limited by the degree to which major persistence is an important
criterion. We propose that major persistence is important because (1) it is an indicator of
satisfaction with academic environment and (2) those who change majors are more likely
to delay their graduation time or drop out. Since the interest-major composite and GPA are
related to major persistence (which we believe is related to satisfaction with entering
major), we hypothesize that both will also be predictive of satisfaction with entering major,
as measured through a reliable battery that measures satisfaction. Future research could test
this hypothesis, as well as the degree to which major persistence is related to satisfaction
with entering major.

Acknowledgment We thank Nancy Petersen, James Sconing, and Kyle Swaney of ACT, Inc. for their
helpful reviews and technical guidance of this study.

References

ACT. (1995). Technical manual: Revised unisex edition of the ACT interest inventory. Iowa City, IA: ACT,
Inc.
ACT. (1997). ACT assessment: Technical manual. Iowa City, IA: ACT, Inc.

123
78 Res High Educ (2008) 49:6279

ACT. (2001). ACTs World-of-Work map revised. Iowa City, IA: ACT, Inc.
Allen, J., Robbins, S., Casillas, A., & Oh, I. (2007). Effects of academic performance, motivation, and social
connectedness on third-year college retention and transfer. Submitted.
Brown, S. P., Gore, P. A. (1994). An evaluation of interest congruence indices: Distribution characteristics
and measurement properties. Journal of Vocational Behavior, 45, 310327.
Classification of Instructional Programs. (2000). (NCES 2002165) US Department of Education National
Center for Education Statistics. Washington, DC: US Government Printing Office.
Dawis, R. L., & Lofquist, L. H. (1984). A psychological theory of work adjustment. Minneapolis, MN:
University of Minnesota Press.
Donohue, R. (2006). Person-environment congruence in relation to career change and career persistence.
Journal of Vocational Behavior, 68, 504515.
Hackett, G., & Lent, R. W. (1992). Theoretical advances and current inquiry in career psychology. In S. D.
Brown &R. W. Lent(Eds.), Handbook of counseling psychology (2nd Edn, pp. 419451). New York:
John Wiley.
Holland, J. L. (1997). Making vocational choices: A theory of vocational personalities and work environ-
ments. (3rd Edn). Odessa, FL: Psychological Assessment Resources.
Kristof-Brown, A. L., Zimmerman, R. D., & Johnson, E. C. (2005). Consequences of individuals fit at
work: A meta-analysis of person-job, person-organization, person-group, and person-supervisor fit.
Personnel Psychology, 58, 281342
Kramer, G.L., Higley, H., & Olsen, D. (1994). Changes in academic major among undergraduate students.
College & University, 69, 8890.
Leuwerke, W. C., Robbins, S., Sawyer, R., & Hovland, M. (2004). Predicting engineering major status from
mathematics achievement and interest congruence. Journal of Career Assessment, 12, 135149.
Lounsbury, J. W., Saudargas, R. A., Gibson, L. W., & Leong, F. T. (2005). An investigation of broad and
narrow personality traits in relation to general and domain-specific life satisfaction of college students.
Research in Higher Education, 46, 707729.
Menard, S. (2000). Coefficients of determination for multiple logistic regression analysis. The American
Statistician, 54, 1724.
Oleski, D., & Subich, L.M. (1996). Congruence and career change in employed adults. Journal of Voca-
tional Behavior, 49, 221229.
Olitsky, N., Robbins, S., & Neumann, G. (2007). Job congruence, academic achievement, and earnings.
Iowa City IA: ACT, Inc. (Manuscript submitted for publication).
Pascarella, E. T., & Terenzini, P. T. (2005). How college affects students: A third decade of research. San
Francisco, CA: Jossey-Bass.
Pike, G. R. (2006). Vocational preferences and college expectations: An extension of Hollands principle of
self-selection. Research in Higher Education, 47, 591612.
Porter, S. R., & Umbach, P. D. (2006). College major choice: An analysis of person-environment fit.
Research in Higher Education, 47, 429449.
Prediger, D. J. (1982). Dimensions underlying Hollands hexagon: Missing link between interests and
occupations? Journal of Vocational Behavior, 21, 259287.
Prediger, D. J., & Swaney, K. B. (2004). Work task dimensions underlying the World-of-Work: Research
results for diverse vocational databases. Journal of Career Assessment, 12, 440459.
Prediger, D. J., & Vansickle, T. R. (1992). Locating occupations on Hollands hexagon: Beyond RIASEC.
Journal of Vocational Behavior, 40, 111128.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models applications and data analysis
methods. Thousand Oaks, CA: Sage Publications.
Robbins, S. B., Allen, J., Casillas, A., Peterson, C., & Le, H. (2006). Unraveling the differential effects of
motivational and skills, social, and self-management measures from traditional predictors of college
outcomes. Journal of Educational Psychology, 98, 598616.
Rounds J. B., & Tracey T. J. (1993). Predigers dimensional representation of Hollands RIASEC cir-
cumplex. Journal of Applied Psychology, 78, 875890.
Slaney, R. (1984). Relations of career indecision to changes in expressed vocational interests. Journal of
Counseling Psychology, 31, 349355.
Smart, J. C., Feldman, K. A., & Ethington, C. A. (2000). Academic disciplines: Hollands theory and the
study of college students and faculty. Nashville, TN: Vanderbilt University Press.
Snyder, T. D., Tan, A. G.,& Hoffman, C. M. (2004). Digtest of education statistics, (NCES 2005-025) US
Department of Education, National Center for Education Statistics. Washington, DC: Government
Printing Office.
Tinto, V. (1993). Leaving college: Rethinking the causes and cures of student attrition (2nd edn). Chicago:
The University of Chicago Press.

123
Res High Educ (2008) 49:6279 79

Tokar, D. M., & Swenson, J. L. (1995). Evaluation of the correspondence between Hollands vocational
personality typology and the five-factor model of personality. Journal of Vocational Behavior, 46,
89108.
Tracey, T. J. G., & Robbins, S. B. (2005). Stability of interests across ethnicity and gender: A longitudinal
examination of grades 8 through 12. Journal of Vocational Behavior, 67, 335364.
Tracey, T. J. G., & Robbins, S. B. (2006). The interest-major congruence and college success relation:
A longitudinal study. Journal of Vocational Behavior, 69, 6489.
Walsh, W. B., & Holland, J. L. (1992). A theory of personality types and work environments. In W. B.
Walsh, K. H. Craik& R. H. Price (Eds.), Person-environment psychology: Models and perspectives
(pp. 3569). Hillsdale, NJ: Lawrence Erlbaum.
Wolfinger, R., & OConnell, M. (1993). Generalized linear mixed models: A pseudo-likelihood approach.
Journal of Statistical Computation and Simulation, 48, 233243.

123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

S-ar putea să vă placă și