corpus.ulaval.ca€¦ · Web viewT1 introjected and external regulations only yielded three...

A Self-Determination Theory Perspective 1

Running Head: Self-Determination Theory and Vocational Interests

A Self-Determination Theory Perspective on RIASEC Occupational Themes: Motivation

Types as Predictors of Self-Efficacy and College Program Domain

Date: March 12, 2019


Abstract

Using the RIASEC (Realistic, Investigative, Artistic, Social, Enterprising,

Conventional; Holland, 1997) model of occupational themes and a one-year prospective

study, we investigated if identified, introjected and external regulations for vocational

activities and their combination are relevant to understand self-efficacy and attendance in

a college program over and above interests (intrinsic motivation). Participants were 966

college students (66% female) who completed measures of motivation types (Time 1) and

self-efficacy (Time 1 and Time 2) toward each RIASEC occupational theme. Results

based on a variable-centered approach revealed that students with autonomous

motivations for a given RIASEC domain generally showed positive changes in self-

efficacy in this domain. Students with high self-efficacy and identified regulation were

also more likely to pursue a program in a corresponding domain. The combination of the

types of motivation to predict outcomes was achieved via person-centered analyses

(latent profile analyses). Results indicated three or four profiles' solution by RIASEC

domain. In general, being in most adaptive profile (high levels of autonomous

motivation, but low levels of controlled motivation) was more adaptive in terms of self-

efficacy or attending a college program than other combination of motivation types.

Results are discussed in light of Self-Determination Theory and the RIASEC model.

Keywords : Vocational Interests, Self-Determination Theory, Vocational Counseling,

Self-Efficacy, Person-Centered.


Vocational interests, conceptualized as preferences for behaviors, situations, and

contexts surrounding work activities (Su, Rounds, & Amstrong, 2009), are recognized as

a motivational strength driving behavior (Nye, Sue, Rounds, & Drasgow, 2012).

According to Holland’s (1997) theory of vocational personalities and work environments,

vocational interests can be categorized into six domains of activities: Realistic (working

with things), Investigative (working with ideas, science-based curiosity), Artistic

(creative expression), Social (working with individuals, helping people), Enterprising

(leadership), and Conventional (structured environment). One central assumption of

Holland’s theory is that individuals whose work fits their vocational interests should be

more satisfied, successful, and dedicated to their work (Nye et al., 2012). Although this

hypothesis has been repeatedly supported (Nauta, 2010), the obtained effect sizes are

small (Tsabari, Tziner, & Meir, 2005). A broader perspective on motivations underlying

vocational activities could help explain why pursuing one’s vocational interests do not

always translate into positive outcomes. Self-determination theory (SDT; Ryan & Deci,

2017) is a promising theoretical framework to carry this investigation.

In SDT, interests are embedded within the concept of intrinsic motivation, a

motivating force based on curiosity, enjoyment, and satisfaction (Ryan & Deci, 2017). It

also conceptualizes other motivating forces that do not stem from the appreciation of an

activity but are rather extrinsic – i.e. carried to achieve an independent goal (Ryan &

Deci, 2017). Extrinsic motivation comprises distinct regulations that represent increasing

levels of self-determination: external, introjected, and identified regulations.

External regulation describes behaviors motivated to obtain a reward or avoid a

punishment. Introjected regulation describes behaviors performed to alleviate internal


pressure (e.g., guilt). External and introjected regulations are considered controlled

motivations, as they are not volitionally endorsed. Identified regulation describes

behaviors that individuals value personally, that are freely chosen. This regulation and

intrinsic motivation are considered autonomous regulations because individuals

voluntarily endorse their behavior. As found in other domains, autonomous regulations

should predict positive vocational outcomes. Here, the focus is on self-efficacy and

domain of college program. Self-efficacy beliefs are relevant because they are positively

associated to choice, performance and persistence (Brown & Lent, 2015). Although some

conceptual models use self-efficacy as a predictor of interests (Lent, Brown, & Hackett,

1994), this study is based on research showing support for the opposite pattern (see

Litalien & Guay, 2015).

Both variable-centered analyses – i.e. estimating relationships between types of

motivation and vocational variables – and person-centered analyses – i.e., identifying

motivational profiles and then comparing vocational variables among these profiles – can

indicate respectively how motivations predict or interact in predicting self-efficacy and

attendance in a college program. For example, in person-centered analyses a group of

students might report high interest (intrinsic motivation) for realistic activities, but also

high levels of identified, introjected, and external regulations for activities in this domain.

Because controlled motivations are strongly endorsed in this profile, SDT would predict

less adaptive vocational outcomes (e.g., lower efficacy) comparatively to a profile where

autonomous motivations are high, but controlled ones are low. Previous studies have

supported the predictive utility of identifying motivational profiles: students show more

persistence when they have high levels of autonomous motivations but low levels of


controlled motivations rather than high levels on all types of motivation (Ratelle et al.,

2007).

Study Goals and Hypotheses

The goal of this study was to predict domain-specific self-efficacy and domain of

college program attendance from SDT’s motivation types for each RIASEC domain,

using variable and person-centered approaches. Four hypotheses (H) were tested:

H1. Intrinsic and identified motivations for a RIASEC dimension at Time 1 (T1) will

positively predict self-efficacy toward activities in this dimension a year later (Time 2;

T2) and attendance in a college program in this dimension at T1 while controlling for the

contribution of T1 self-efficacy;

H2. Introjected and external regulations for a RIASEC dimension at T1 will negatively

predict self-efficacy toward activities in this dimension at T2 and attendance in a study

program in this dimension at T1 while controlling for T1 self-efficacy;

H3. Various profiles will emerge within a specific RIASEC dimension: an

autonomous profile where autonomous motivations are high but controlled ones are low;

a high profile where all types of motivation are high; a low profile were all types of

motivation are low; and a controlled profile where controlled motivations are high but

autonomous ones are low.

H4. Within each RIASEC domain the autonomous profiles will evidence the highest

level of self-efficacy and the greatest proportion of attendance in a college program in the

corresponding dimension comparatively to the other profiles.

MethodParticipants and Procedure1


Data comes from a one-year longitudinal study on college students’ vocational

decision-making (Poitras, Guay & Ratelle, 2012). A total of 966 students (66% female)

complete a questionnaire on site at T1 and 332 of them completed a online questionnaire

at T2 (72% female). The target sample size was 1000 participants, which would provide

sufficient statistical power for complex modeling. Participant recruitment was stopped at

976 students, which was very close to the initial goal.

Measures1

The Activities section of the French-Canadian reduced SDS scale (Holland, 1991)

was used to assess students’ motivations toward the RIASEC domains (96 items : 4

motivations X 24 activities assessing the 6 RIASEC domains; Poitras, Guay & Ratelle,

2012). In the T1 questionnaire, participants were asked, following a general question that

prompted answers relative to a specific motivation type (e.g., for identified regulation: “I

would do the following activities because I find them important”) how motivated they

were regarding each activity (24 activities in total). This procedure was repeated for

intrinsic motivation as well as identified, introjected, and external regulations. The

response format ranged from 1 to 5 for all measures. Self-efficacy toward RIASEC

activities (n = 24) was assessed at T1 and T2. McDonald’s (1970) omega for all measures

ranged from .75 to .93 across variables and vocational domains. Participants study

programs at T1 were classified according to RIASEC dimensions to represent program

attendance. Participants in the sample were studying in three main domains: investigative

(N=189), artistic (N=71), and social (N=428).

Results1

Variable-Centered Analyses (Hypotheses 1 and 2)


Structural equation modeling (Mplus 8; Muthén & Muthén, 2017), was used to test

models within RIASEC domains in which motivation types predicted self-efficacy and

program attendance in the corresponding domain (6 models for the prediction of self-

efficacy and 3 for the prediction of program attendance; fit indices are presented in Table

S2). Results are presented in Table 1. Results indicated that T1 intrinsic motivation

positively predicted T2 self-efficacy in the corresponding dimension controlling for T1

within artistic (.45) and realistic (.27) domains. T1 identified regulation also positively

predicted T2 self-efficacy in the corresponding dimension, within the social (.28) domain.

T1 introjected and external regulations only yielded three statistically significant paths in

corresponding domains, two in the conventional domain, where self-efficacy was

negatively and weakly predicted by introjected regulation (-.10) and positively predicted

by external regulation (.11), and one in the investigative domain, where self-efficacy was

negatively and weakly predicted by introjected regulation (-.09). These results held while

controlling for the contribution of T1 self-efficacy with domains.

In models predicting within-domain program attendance, models could only be

estimated for investigative, artistic, and social domains (Table 1). Intrinsic motivation

was not associated with attending a corresponding program in the social domain (.16; not

statistically significant), although the coefficient was in the positive direction and

represents a small effect size. The opposite was found for intrinsic motivation for an

artistic domain, which negatively predicted program attendance (-.19), possibly reflecting

a suppression effect. Identified regulation was linked with attending a corresponding

program in investigative (.23) and artistic (.26) domains. Only one domain found external

regulation to predict program attendance, the social one (-.09). Self-efficacy at T1 was


associated with attending a corresponding program in the three domains. H1 and H2 were

thus partially supported.

Person-centered approach (Hypotheses 3 and 4)

A three-profile solution was identified as optimal in most vocational domains, apart

from the social domain (4 profiles). Two profiles were similar across all domains, while

two other profiles were only shared across specific domains. The final profile solution

within each domain is depicted in Figure 1. In each domain, a “Autonomous” profile was

identified and included students reporting high levels of intrinsic motivation and

identified regulation and low levels of introjected and external regulations within

domains. A second profile was labeled “Low Motivation” and included students whose

scores on all types of motivation were below average. A third profile differed across

RIASEC domains, with two recurring patterns. A first pattern pertained to R, A, S, and C

domains and was labelled “Introjected” because this type of motivation was higher than

all the others, which also remained above the mean. The second pattern, observed across

I, S, and E domains was labelled “Moderate” as all types of regulations were close to the

mean. Finally, for the S domain, we observed the four profiles described above. Intrinsic

motivation scores were above the mean for both Autonomous and Introjected profiles,

implying that students’ level of vocational interest (intrinsic motivation) is not

sufficiently informative to discriminate these profiles.

To test the validity and utility of these profiles, auxiliary analyses were performed

within Mplus to compare profiles on T1 and T2 levels of corresponding self-efficacy and

T1 program domain attendance (Table 2). Within each RIASEC dimension, there were

statistically significant differences in corresponding self-efficacy at T1 and T2 across


profiles, except for the conventional and investigative domains. Specifically, the

Autonomous profile appeared to be the most adaptive one, with students reporting high

levels of self-efficacy as well as attending a college program that corresponded to the

target domain. Furthermore, although the Introjected profile found in RAC domains was

characterized by moderately high levels of intrinsic motivation, being in this profile or in

the Moderate one was associated with lower self-efficacy than being in the Autonomous

profile. These profiles also discriminated program attendance, except in the artistic

domain where only one statistically significant difference was observed across the

profiles. However, the small number of participants pursuing an artistic program (N = 71)

might explain such undifferentiated results. Finally, the least adaptive profile in each

RIASEC domain was the Low Motivation profile, whose participants reported the lowest

levels of corresponding self-efficacy, compared to those in other profiles. This profile

also predicted attending a program in another domain. These results therefore confirmed

H3 and H4.

Discussion

Interests are widely used to guide vocational choice. The current findings indicate

that they might be insufficient to accurately predict self-efficacy in a vocational domain.

Indeed, other motivational components within RIASEC activity domains also need to be

considered, namely whether domain activities are perceived as personally important

(identified regulation) or contingent upon internal (introjected regulation) or external

pressures (external regulation). However, levels of interests did discriminate college

program domain because no difference was found between the Autonomous and

Introjected profile, the only pair of profiles that showed interest levels above the mean.


The findings concur with those of previous studies showing that autonomous

motivations predict positive educational (Guay et al., 2008) and vocational outcomes

(Guay et al., 2006). One peculiar finding relates to artistic activities, where intrinsic

motivation positively predicted changes in self-efficacy but correlated negatively with

attending an artistic college program. Identified regulation was, in contrast, positively

associated with attending an artistic college program. Future studies should try to

understand this finding specific to the art domain. Choosing to pursue a domain of

activity and feeling efficient within this domain appears to require more than mere

interest. Rather, it needs to be based on more stable motivational anchoring such as

values and importance, which are less likely to fluctuate when setbacks occur (Vallerand,

Gauvin, & Halliwell, 1986).

The contributions of controlled regulations appear to be more indirect as results

from SEM analyses provide little support for their relevance. Specifically, controlled

regulations might moderated the contribution of autonomous motivations on vocational

adjustment such that profiles with highly controlled regulations reduce the prediction of

adaptive vocational outcomes by autonomous motivations. A more accurate prediction of

students’ vocational adjustment within domain appears to require considering how

autonomous and controlled motivations combine within domain instead of examining

these two categories of motivational factors in isolation. In other words, not all vocational

interests are created equal: those accompanied with low levels of controlled regulations

appear more beneficial for vocational adjustment.

Because interests are only one of the important motivational mechanisms

underlying vocational adjustment, vocational counselors would benefit from taking into


account their clients’ other motivations when examining their interest for each domain of

activity. Doing so should benefit their clients’ vocational adjustment. Hence, when

students report high levels of interest for multiple domains on the RIASEC, vocational

counselors can verify if students value these or other domains (i.e., whether they have an

identified regulation), and if they experience internal (e.g., self-esteem contingencies;

introjected regulation) or external (e.g., income incentives; external regulation) pressures

to consider these domains. In the latter cases, the current findings suggest that student

pursuing a domain where interests are combined with high identified but low controlled

regulations would be more adapted.

Despite this study’s meaningful results, some shortcomings need to be outlined.

First, while a 12-month prospective design was used, it is inappropriate to infer causal

relations since the design was descriptive. Second, only the T1 measure of the domain of

college program was used because the T2 measure had too many missing data. Third,

since the domain of a program could only be classified in three RIASEC dimensions

(investigative, artistic, and social), replicating these findings with programs in other

domains will be important.


Footnote

1. More details are provided in the online supplementary materials.

References

Bauer, D. J., & Curran, P. J. (2004). The integration of continuous and discrete latent

variable models: Potential problems and promising opportunities. Psychological

methods, 9, 3-29. dx.doi.org/10.1037/1082-989X.9.1.3

Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K.A.

Bollen & J. Scott Long (Eds.), Testing Structural Equation Models (pp. 136-154).

London, UK: Sage focus editions.

Brown, S. D., & Lent, R. W. (2015). Social cognitive career theory: A theory of self

(efficacy) in context. In F. Guay, H.W. Marsh, D.M. McInerney, & R. Craven,

Self-concept, motivation, and identity: Underpinning success with research and

practice (173-200). Charlotte, NC: Information Age Publishing.

Byrne, B. M. (2013). Structural equation modeling with Mplus: Basic concepts,

applications, and programming. New York, NY: Routledge.

Davey, A., Shanahan, M. J., & Schafer, J. L. (2001). Correcting for selective nonresponse

in the National Longitudinal Survey of Youth using multiple imputation. Journal

of Human Resources, 36, 500-519. DOI: 10.2307/3069628

Enders, C.K. (2010). Applied missing data analysis. New York: Guilford.

Graham, J.W. (2009). Missing data analysis: Making it work in the real world. Annual

Review of Psychology, 60, 549-576.

Doi.org/10.1146/annurev.psych.58.110405.085530.

Guay, F., Ratelle, C. F., Senécal, C., Larose, S., & Deschênes, A. (2006). Distinguishing

developmental from chronic career indecision: Self-efficacy, autonomy, and

social support. Journal of Career Assessment, 14, 235-251.

doi.org/10.1177/1069072705283975

https://doi.org/10.1177%2F1069072705283975

http://psycnet.apa.org/doi/10.1037/1082-989X.9.1.3


Guay, F., Ratelle, C. F., & Chanal, J. (2008). Optimal learning in optimal contexts: The

role of self-determination in education. Canadian Psychology/Psychologie

canadienne, 49, 233-240. dx.doi.org/10.1037/a0012758

Hipp, J. R., & Bauer, D. J. (2006). Local solutions in the estimation of growth mixture

models. Psychological Methods, 11, 36-53. dx.doi.org/10.1037/1082-

989X.11.1.36

Hogan, T. P. (2007). Psychological testing: a practical introduction (2nd ed.). Hoboken,

NJ: John Wiley.

Holland, J. L. (1991). L’orientation par soi-même : édition canadienne-française.

Odessa, FL: Psychological Assessment Resources.

Holland, J. L. (1997). Making vocational choices: A theory of vocational personalities

and work environments (3rd ed.). Odessa, FL: Psychological Assessment

Resources.

Howard, J., Gagné, M., Morin, A. J., & Van den Broeck, A. (2016). Motivation profiles

at work: A self-determination theory approach. Journal of Vocational Behavior,

95, 74-89. doi.org/10.1016/j.jvb.2016.07.004

Lent, R. W., Brown, S. D. & Hackett, G. (1994). Toward a unifying social cognitive

theory of career and academic interest, choice, and performance. Journal of

Vocational Behavior, 45, 79-122. doi.org/10.1006/jvbe.1994.1027

Litalien, D., & Guay, F. (2015). Dropout intentions in PhD studies: A comprehensive

model based on interpersonal relationships and motivational

resources. Contemporary Educational Psychology, 41, 218-231.

doi.org/10.1016/j.cedpsych.2015.03.004

Lubke, G., & Muthén, B. O. (2007). Performance of factor mixture models as a function

of model size, covariate effects, and class-specific parameters. Structural

Equation Modeling, 14, 26-47.

https://doi.org/10.1016/j.cedpsych.2015.03.004

https://doi.org/10.1006/jvbe.1994.1027

https://doi.org/10.1016/j.jvb.2016.07.004



http://psycnet.apa.org/doi/10.1037/a0012758


Marsh, H. W., Lüdtke, O., Trautwein, U., & Morin, A. J. (2009). Classical latent profile

analysis of academic self-concept dimensions: Synergy of person-and variable-

centered approaches to theoretical models of self-concept. Structural Equation

Modeling, 16, 191-225. doi.org/10.1080/10705510902751010

McDonald, R. P. (1970). The theoretical foundations of principal factor analysis,

canonical factor analysis, and alpha factor analysis. British Journal of

Mathematical and Statistical Psychology, 23, 1-21. doi.org/10.1111/j.2044-

8317.1970.tb00432.x

McLachlan, G., & Peel, D. (2000). Multivariate normal mixtures. Finite Mixture Models,

New York : Wiley, 81-116.

Morin, A. J., Maiano, C., Nagengast, B., Marsh, H. W., Morizot, J., & Janosz, M. (2011).

General growth mixture analysis of adolescents' developmental trajectories of

anxiety: The impact of untested invariance assumptions on substantive

interpretations. Structural Equation Modeling: A Multidisciplinary Journal, 18,

613-648. doi.org/10.1080/10705511.2011.607714

Muthén, B. O. (2002). Beyond SEM: General latent variable modeling.

Behaviormetrika, 29, 81-117. doi.org/10.2333/bhmk.29.81

Muthén, B. (2003). Statistical and substantive checking in growth mixture modeling:

comment on Bauer and Curran. Psychological Methods, 8, 369-77; discussion

384-93. doi.org/10.1037/1082-989X.8.3.369

Muthén, L. K., & Muthén, B. O. (2017). Mplus user’s guide. Eighth Edition. (Eighth

Ed.). Los Angeles, CA: Muthén & Muthén.

Nauta, M. M. (2010). The development, evolution, and status of Holland’s theory of

vocational personalities: Reflections and future directions for counseling

psychology. Journal of Counseling Psychology, 57, 11-22.

dx.doi.org/10.1037/a0018213



https://www.ncbi.nlm.nih.gov/pubmed/14596497

https://doi.org/10.2333/bhmk.29.81

https://doi.org/10.1080/10705511.2011.607714

https://doi.org/10.1111/j.2044-8317.1970.tb00432.x


https://doi.org/10.1080/10705510902751010


Nye, C. D., Su, R., Rounds, J., & Drasgow, F. (2012). Vocational interests and

performance: A quantitative summary of over 60 years of research. Perspectives

on Psychological Science, 7, 384-403. doi.org/10.1177/1745691612449021

Peugh, J., & Fan, X. (2013). Modeling unobserved heterogeneity using latent profile

analysis: A Monte Carlo simulation. Structural Equation Modeling: A

Multidisciplinary Journal, 20, 616-639. doi.org/10.1080/10705511.2013.824780

Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common

method biases in behavioral research: A critical review of the literature and

recommended remedies. Journal of applied psychology, 88, 879-903.

doi:10.1037/0021-9010.88.5.879

Poitras, S-C., Guay, F., & Ratelle, C. F. (2012). Using the self-directed search in

research: Selecting a representative pool of items to measure vocational interests.

Journal of Career Development, 39, 186-207. doi:10.1177/0894845310384593

Ratelle, C. F., Guay, F., Vallerand, R. J., Larose, S., & Senécal, C. (2007). Autonomous,

controlled, and amotivated types of academic motivation: A person-oriented

analysis. Journal of educational psychology, 99, 734-736.

dx.doi.org/10.1037/0022-0663.99.4.734

Ryan, R. M., & Deci, E. L. (2017). Self-determination theory: Basic psychological needs

in motivation, development, and wellness. New York, NY: Guilford Publications.

Schumacker, E. R., & Lomax, G. R. (1996). A beginner’s guide to structural equation

modeling. Mahwah, NJ: Lawrence Erlbaum.

Su, R., Rounds, J., & Armstrong, P. I. (2009). Men and things, women and people: a

meta-analysis of sex differences in interests. Psychological Bulletin, 135, 859-

884. dx.doi.org/10.1037/a0017364

Tsabari, O., Tziner, A., & Meir, E. I. (2005). Updated meta-analysis on the relationship

between congruence and satisfaction. Journal of Career Assessment, 13, 216-232.

doi.org/10.1177/1069072704273165

https://doi.org/10.1177%2F1069072704273165


http://psycnet.apa.org/doi/10.1037/0022-0663.99.4.734

http://www.fse.ulaval.ca/cv/catherine.ratelle/publications/Using%20the%20self-directed%20search%20in%20research:%20Selecting%20a%20representative%20pool%20of%20items%20to%20measure%20vocational%20interests.%20Journal%20of%20Career%20Development,%2039(2),%20186-207.http:/dx.doi.org/10.1177/0894845310384593

https://doi.org/10.1080/10705511.2013.824780

https://doi.org/10.1177%2F1745691612449021


Vallerand, R. J., Gauvin, L. I., & Halliwell, W. R. (1986). Effects of zero-sum

competition on children's intrinsic motivation and perceived competence. The

Journal of Social Psychology, 126(4), 465-472.

doi.org/10.1080/00224545.1986.9713614

Yu, C. Y. (2002). Evaluation of model fit indices for latent variable models with

categorical and continuous outcomes. Paper presented at the Annual Conference

of the American Educational Research Association, April 2002, New Orleans.

https://doi.org/10.1080/00224545.1986.9713614

Supplemental Online Material 17

Table 1 Results from the variable-centered analyses: Factor loadings and path coefficients.Type of domain Mean L L-SD L-range β for the prediction

of T2 self-efficacy ΔR2 β for the prediction

of T1 program ΔR2

RealisticSelf-efficacy T1 .82 .06 .73-.86 .60* .60* ---Intrinsic .85 .06 .77-.91 .27*

.04*

---Identified .86 .06 .78-.92 .00 ---Introjected .88 .05 .83-.93 -.08 ---External .85 .05 .79-.90 .00 ---InvestigativeSelf-efficacy T1 .81 .04 .77-.86 .73* .66* .33* .31*Intrinsic .82 .05 .75-.87 .10

.02*

.08

.02*Identified .82 .04 .79-.86 .02 .23*Introjected .85 .03 .81-.88 -.09* -.04External .80 .05 .76-.85 .05 .00ArtisticSelf-efficacy T1 .66 .15 .48-.81 .47* .62* .32* .13*Intrinsic .71 .17 .53-.89 .45*

.12*

-.19*

.02*Identified .74 .14 .58-.87 -.00 .26*Introjected .82 .08 .73-.90 -.00 -.08External .70 .14 .56-.84 -.01 .06SocialSelf-efficacy T1 .57-.77 .68* .61* .32* .13*Intrinsic .71 .09 .61-.82 -.07

.10*

.16

.02*Identified .70 .09 .60-.82 .28* -.12Introjected .84 .05 .79-.90 -.03 .05External .71 .08 .62-.82 -.08 -.09*EnterprisingSelf-efficacy T1 .67 .08 .66-.82 .65* .50* ---Intrinsic .80 .06 .73-.85 .02

.04*

---Identified .80 .06 .74-.86 .15 ---Introjected .86 .04 .83-.91 -.03 ---External .79 .07 .69-.85 -.04 ---ConventionalSelf-efficacy T1 .84 .06 .76-.88 .70* .51* ---Intrinsic .86 .03 .83-.89 .06

.02*

---Identified .83 .04 .78-.86 -.03 ---Introjected .88 .02 .86-.92 -.10* ---External .82 .04 .77-.85 .11* ---Note. * p < .05. T1 = Time 1; T2 = Time 2; SD = standard deviation; L = loadings. This table presents results from nine sets of SEM models, each set first testing the prediction of the outcome (latent T2 self-efficacy or T1 program) by T1 self-efficacy, followed by another model testing the prediction of the outcome by five T1 exogenous latent variables (intrinsic motivation, identified, introjected, and external regulations, and self-efficacy). The percentage of R2 attributable to T1 self-efficacy and the incremental prediction by SDT constructs are presented. All R2 are significant at *p < .05.


Table 2 Means and Standard Errors for Each Variable PerProfile Within Each RIASEC Domains

Means by profiles (SE) Ms (SE) for entire sampleAutonomous

profileIntrojected

profileModerate

profileLow profile

Realistic n=244 n=135 n = 595 n=974 Intrinsic 1.85 (0.22)a 0.97 (0.18)b - -0.99 (0.07)c 0.00 (0.03) Identified 1.36 (0.12)a 1.12 (0.16)a - -0.82 (0.09)b 0.00 (0.03) Introjected -0.6 (0.10)a 4.29 (0.22)b - -0.74 (0.07)a 0.00 (0.02) External regulation 0.03 (0.06)]a 0.66 (0.08)b - -0.16 (0.05)a 0.00 (0.04) Self-efficacy T1 0.61 (0.05)a 0.49 (.06)a -0.39 (0.02)b 0.00 (0.03) Self-efficacy T2 0.84 (0.08)a 0.16 (0.11)b -0.42 (0.04)c 0.00 (0.03)Investigative n=153 n=291 n=530 n=974 Intrinsic 4.48 (0.35)a - 1.07 (0.27)b -1.91 (0.11)c 0.00 (0.03) Identified 2.57 (0.19)a - 0.97 (0.09)b -1.29 (0.14)c 0.00 (0.04) Introjected 0.36 (0.12)a - 0.31 (0.07)a -0.28 (0.05)b 0.00 (0.03) External regulation -0.10 (0.09)a - 0.24 (0.05)b -0.11 (0.05)a 0.00 (0.04) Self-efficacy T1 1.26 (0.05)a 0.47 (0.05)b -0.79 (0.04)c 0.00 (0.03) Self-efficacy T2 1.04 (0.09)a 0.27 (0.09)b -0.53 (0.07)c 0.00 (0.03) Program 0.66 (0.05)a - 0.43 (0.04)b 0.08 (0.02)c 0.27 (0.02)Artistic n=251 n=146 n=577 n=974 Intrinsic 2.08 (0.11)a 0.79 (0.29)b - -1.13 (0.08)c 0.00 (0.04) Identified 1.64 (0.12)a 0.98 (0.24)a - -0.98 (0.05)b 0.00 (0.04) Introjected -0.82 (0.10)a 3.82 (0.18)b - -0.60 (0.15)a 0.00 (0.03) External regulation -0.11 (0.07)a 0.48 (0.09)b - -0.07 (0.05)a 0.00 (0.04) Self-efficacy T1 0.88 (0.05)a 0.41 (0.07)b -0.52 (0.03)c 0.00 (0.03) Self-efficacy T2 0.84 (0.08)a 0.11 (0.12)b -0.50 (0.06)c 0.00 (0.03) Program .19 (0.03)ac .11 (0.03)bc - 0.06 (0.01)b 0.10 (0.01)Social n=244 n=147 n=357 n=226 n=974 Intrinsic 2.38 (0.14)a 1.80 (0.22)b -0.42 (0.12)c -3.06 (0.12)d 0.00 (0.03) Identified 2.29 (0.11)a 1.97 (0.21)a -0.44 (0.11)b -3.03 (0.15)c 0.00 (0.02) Introjected -0.94 (0.08)a 3.03 (0.18)b -0.22 (0.13)c -0.59 (0.10)c 0.00 (0.03) External regulation -0.31(0.09)a 0.97 (0.10)b -0.03 (0.05)c -0.23 (0.07)a 0.00 (0.03) Self-efficacy T1 0.50 (0.02)a 0.37 (0.03)b -0.09 (0.02)c -0.63 (0.03)d 0.00 (0.03) Self-efficacy T2 0.50 (0.02)a 0.33 (0.08)a -0.06 (0.05)b -0.68 (0.08)c 0.00 (0.03) Program 0.78 (0.03)a 0.74 (0.04)a 0.59 (0.03)b 0.41 (0.04)c 0.62 (0.02)Enterprising n=247 n=409 n=318 n=974 Intrinsic 2.81 (0.24)a - 0.05 (0.20)b -2.26 (0.10)c 0.00 (0.04) Identified 2.59 (0.12)a - 0.26 (0.17)b -2.36 (0.23)c 0.00 (0.04) Introjected 0.47 (0.10)a - 0.01 (0.05)b -0.38 (0.05)c 0.00 (0.03) External regulation 0.28 (0.09)a - 0.06 (0.04)a -0.30 (0.07)b 0.00 (0.03) Self-efficacy T1 0.85 (0.04)a 0.06 (0.03)b -0.71 (0.04)c 0.00 (0.03) Self-efficacy T2 0.61 (0.06)a 0.02 (0.07)b -0.59 (0.08)c 0.00 (0.03)Conventional n=214 n=182 n=578 n=974 Intrinsic 1.75 (0.18)a 1.00 (0.16)b - -0.98 (0.06)c 0.00 (0.04) Identified 1.53 (0.09)a 0.88 (0.13)b - -0.85 (0.08)c 0.00 (0.04) Introjected -0.87 (0.09)a 3.50 (0.14)b - -0.76 (0.10)a 0.00 (0.03) External regulation 0.10 (0.07)a 0.46 (0.08)b - -0.18 (0.05)c 0.00 (0.03) Self-efficacy T1 0.89 (0.05)a 0.49 (0.07)b -0.58 (0.04)c 0.00 (0.03) Self-efficacy T2 0.76 (0.10)a 0.15 (0.13)b -0.31 (0.07)c 0.00 (0.03)

Note. T1 = Time 1; T2 = Time 2; SE = standard error. All scores, except for college program attendance, are standardized (M = 0, SD = 1). To interpret mean differences across profiles, readers should compare scores horizontally across profiles for a given RIASEC domain, not vertically within a profile nor across RIASEC domains. Different letters mean statistically significant differences across profiles at p < .05. For example, for T2 self-efficacy in Realistic activities, there are differences between Autonomous and Introjected profiles, as well as between the Autonomous and Low profiles.


Figure 1. Scores for Each Type of Motivation in a Given Profile Within Each RIASEC Domain

-2

-1

0

1

2

3

4

5

Realistic

Intrinsic Identified Introjected External Série5

autonomousprofile 25%

introjectedprofile 14%

low profile61%

-3

-2

-1

0

1

2

3

4

5

Investigative

Intrinsic Identified Introjected External


moderateprofile 30%

low profile54%

-2

-1

0

1

2

3

4

5

Artistic




low profile59%

-4

-3

-2

-1

0

1

2

3

4SOCIAL



moderatedprofile 37%

low profile 23%


-3

-2

-1

0

1

2

3

4

Enterprising



moderateprofile 42%

low profile33%

-1,5-1

-0,50

0,51

1,52

2,53

3,54

Conventional




low profile59%

Note. This figure depicts standardized motivation type scores for each profile within each RIASEC domain. The number of students in each profile is expressed in percentage. There are three or four profiles by RIASEC domain. Two profiles are consistent across all 6 domains (Autonomous and low profiles). RASC domains have an introjected profile and ISE domains have a moderate profile.


Supplemental Online Material for:

A Self-Determination Theory Perspective on RIASEC Activities Types: Motivation

Types as Predictors of Self-Efficacy and College Program Domain


Complete Methodology

Participants

The institutional review board approved this project. Data comes from a one-year

longitudinal study on college students’ vocational decision-making (Poitras, Guay &

Ratelle, 2012), partly carried to validate a shortened version of the Self-Directed Search

used to assess interests for RIASEC domains. The study published in 2012 focused on

vocational interests exclusively whereas, in this study, self-efficacy and types of

motivation for each RIASEC domain were used. In the fall of 2007 (Time 1; T1),

research assistants visited libraries, student cafés, and cafeterias of 11 colleges in the

province of Quebec to ask students to fill a questionnaire that included measures of

vocational motivations and other related constructs. Nine hundred seventy-six students

participated in the first wave of this longitudinal study by completing a paper

questionnaire on site (66% female; 2% unspecified). Their mean age was 18.85 years (SD

= 2.59) and 39% of them were in their first college semester. Most of them (98%) spoke

French at home and more than half lived with both of their parents (55%). With regard to

parental levels of education, 32% of mothers earned a high school diploma, 27% a

college diploma, and 31% a university diploma (10% unspecified). For fathers, 29% of

them earned a high school diploma, 21% a college diploma, and 35% a university

diploma (15% unspecified). The target sample size was 1000 participants, which would

provide sufficient statistical power for complex modeling. Participant recruitment was

stopped at 976 students, which was very close to the initial goal.

In September 2008, participants who completed a questionnaire at T1 (N = 976)

were contacted by phone and asked to complete an online survey that included the


domain-specific self-efficacy measure. Participants who completed the Time 2 (T2)

questionnaire were eligible to win one of 25 pairs of movie tickets. Participants who

consented received personalized login information to access the online survey. Up to

three additional phone calls were made to remind participants to complete the survey. A

total of 332 students participated at T2 (72% female). Because 66% of students were

missing at T2, we verified if those who completed both data waves were equivalent to

those who only participated at T1.

Measures

Motivation types. The Activities section of the French-Canadian version of the Self-

Descriptive Search (SDS; Holland, 1991) was used to assess students’ motivations

toward vocational domains of the RIASEC. In the original scale assessing vocational

interests, participants indicated whether they would be interested in engaging in each of

the 66 proposed career-related activities (11 per RIASEC dimension). In this study, we

used a short version of the SDS (24 items, 4 activities per RIASEC dimension; Poitras,

Guay & Ratelle, 2012) and added identified, introjected, and external regulations for each

of the 24 activities. The decision not to use the full version of the SDS was based on the

excessive length the questionnaire would take if all activities were surveyed (see below

for more details). This shortened version was obtained using item response theory and

confirmatory factor analyses (CFA; Poitras, Guay & Ratelle, 2012). Sample activities are

painting (Artistic), teaching (Social), and being a salesperson (Enterprising). The

questionnaire assessing the four types of motivation for all activity domains began by a

statement reflective a specific type of motivation, and asked participants to indicate the

extent to which each domain activity would be done for this motivation. For example, the


intrinsic motivation subscale would state “I would do the following activities by

pleasure” and the 24 activities (e.g., painting, teaching, being a sale person) of the short

version of the SDS would be answered with respect to this statement. Students indicated

on a scale ranging from 1 (not at all agree) to 5 (totally agree) if they would engage in

each activity for this specific reason. After completing all 24 items for intrinsic

motivation, students were asked to complete the same 24 items, but for identified

regulation (“I would do the following activities because I find them important”),

introjected regulation (“I would do the following activities to show others that I’m

capable”), and external regulation (“I would do the following activities only if I receive

something in return”). Ninety-six items were thus used to assess the four types of

motivation toward all RIASEC dimensions (4 motivation statements X 24 activities = 96

items). In other words, the group of 24 SDS items are presented to participants 4 times,

but under a different general motivation statement assessing intrinsic, identified,

introjected, and external regulation. McDonald’s (1970) omega for each type of

motivation ranged from .80 to .93 across the six vocational domains. In the original

version of the SDS, participants answered according to a Yes/No format. In order to

increase variability in participants’ responses, which can increase reliability estimates

(Hogan, 2007), we modified the response scale – with the approval of Psychological

Assessment Resources, Inc. – to a 5-point Likert scale ranging from 1 to 5. This response

format can also improve fit indices in subsequent analyses, such as CFA.

Self-efficacy. Domain-specific self-efficacy toward RIASEC activities was

assessed at T1 and T2 with the statement “Indicate your confidence level regarding your

ability to execute the following activities”, which was answered for each of the 24


activities of the short version of the SDS. The metric used was the same as for types of

motivation. Across vocational domains, omega values for self-efficacy scores ranged

from .75 and .91 at T1 and from .75 to .93 at T2.

College program attendance. The programs in which participants were studying

at T1 were classified according to RIASEC dimensions. Each program could be classified

in one of three main domains: investigative (N = 189), artistic (N = 71), and social (N =

428). Unfortunately, too few students in the sample (< 3%) pursued a college program

that could be classified in realistic, conventional, or enterprising domains. Thus, only

three domains out of six were considered for college program attendance. This repartition

is an artefact of the data collection procedure, which occurred in colleges, which have

three main pre-university programs: natural sciences (classified as Investigative), arts and

languages (classified as Artistic), and social sciences and humanities (classified as

Social). These educational institutions also have technical degrees in Realistic domains,

but these curriculums have a much tighter schedule, meaning that their students were less

accessible at the time of data collection. In addition, it is very likely that students in social

sciences and humanities go on to pursue a university degree in an Enterprising or

Conventional domain as their main domain of interest. However, at this level of

instruction in the Province of Quebec, it was impossible to distinguish these students

from those only interested in the Social domain.

Statistical Analyses

Structural Equation Modeling (SEM)

Additional information on tested SEM models . Because the motivation and

self-efficacy measures used different defining statements but the activities (i.e., items)


was constant across all types of motivation and self-efficacy (the same 24 items were

used), correlations between latent factors might be inflated due to shared wording of

items, or the general model fit might decrease because of strongly related residuals. For

example, students had to indicate the extent to which they find painting to be an

intrinsically motivating activity as well as if they identified with it, if they had introjected

or external regulations for doing it, and their level of self-efficacy toward it. Hence, five

distinct items included the same wording to describe a specific activity (e.g., the word

“painting” was used for the following items: Intrinsic_Act1_A, Identified_Act1_A,

Introjected_Act1_A, External_Act1_A, and Self-efficacy_ Act1_A). To minimize

statistical closeness due to wording artefacts, correlated uniquenesses were estimated

between each items at T1 referring to the same activity (40 correlated uniquenesses = 10

correlations x 4 activities). Moreover, longitudinal correlated uniquenesses were

estimated, but only for self-efficacy error terms (same items used at both time

measurement). Because the CFA and SEM models tested in this study already involved

many free parameters, the correlated uniquenesses models appeared most appropriate

compared to other models used to control potential bias (e.g., multiple-method factor

approach; Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). In addition to all the

statistical parameters (i.e., factor loadings, correlated uniquenesses) described for the

measurement CFA models, SEM models included structural paths estimated in

accordance with the proposed hypotheses. In Figure S1, a SEM model illustrates the

parameters estimated in models presented in the manuscript, where types of motivation

predicted T2 self-efficacy, controlling for T1 self-efficacy, including all correlated

uniquenesses. For SEM models predicting college program attendance, the endogenous


(i.e., T2 self-efficacy) factor was replaced by the dichotomous variable representing

attending a program in the corresponding domain. Longitudinal correlated uniquenesses

were thus removed in these models.

To ascertain the adequacy of model fit of CFA and SEM models, we used the

comparative fit index (CFI), Tucker-Lewis index (TLI), root mean square error of

approximation (RMSEA), the standardized root mean square residual (SRMR), and the S-

B χ2 statistic. CFI and TLI usually vary along a 0 -1 continuum (the TLI can be greater

than 1, which could imply overfitting) and values greater than .90 and .95 typically reflect

acceptable and excellent fit to the data, respectively (Schumacker & Lomax, 1996).

Browne and Cudeck (1993) suggested that RMSEA values below .05 are indicative of a

“close fit” and that values of up to .08 represent reasonable errors of approximation.

Similar guidelines were formulated for the SRMR (Byrne, 2013). Whereas the TLI and

RMSEA penalize lack of parsimony, the CFI and SRMR do not. Under the WLSMV

estimator, the weighted root-mean-square residual (WRMR) will be reported instead of

the SRMR, with values below 1 indicating a satisfactory fit (Yu, 2002). Correlations

among latent factors within each vocational domain are presented in Table S1, while

Table S2 presents fit indices for the various models presented in the manuscript as well as

for all additional models (see below).

Additional SEM models. Three series of models were estimated but could not be

presented in the main manuscript because of space restrictions. First are CFA models that

allowed the evaluation of measurement adequacy of latent factors (Models S1, Table S2;

6 models). Second, SEM models were estimated in which motivations in one domain

predicted self-efficacy in each domain (Models S2; 6 models). A final set of models


included motivations in one domain predicting attending a program in an investigative,

artistic, or social domain (Models S3; 6 models). These 18 additional models are

different from the 9 models presented in the main manuscript as they each include many

more dependent variables, making it possible to explore cross-domain results. These

various models were also estimated using Mplus (Version 8; Muthén & Muthén, 2017)

with the robust maximum likelihood (MLR) estimation method. Because models in the

main manuscript include self-efficacy at T1 as a control variable, they could present

suppression effects. For this reason, Tables S4 and S5 present results based on SEM

analyses without T1 self-efficacy as a covariate. They also present regression coefficients

between non-corresponding domains.

Latent profile analyses. Latent profile analyses (LPA; Muthén, 2002) were

conducted to identify motivational profiles within domains. Solutions with 1 to 8 profiles

were examined separately within each domain. These analyses were performed using

factor scores saved from the CFA models. These factor scores, specified to have a mean

of 0 and a standard deviation of 1, reflect participants’ levels on each of the four latent

factors representing types of motivation (intrinsic motivation and identified, introjected,

and external regulations) for each domain. In the LPA, the means for each type of

motivation were freely estimated (Peugh & Fan, 2013; Morin et al., 2011) and models

were tested with both free and fixed variances on indicators between profiles. Models

were estimated using 5000 random sets of start values, with 100 iterations for each

random start. The 200 best solutions were retained for final stage optimization (Hipp &

Bauer, 2006; McLachlan & Peel, 2000). Results for all profile analyses are presented in

Table S5.


Determining the optimal number of profiles should rely on the theoretical conformity

of the obtained profiles (Marsh, Lüdtke, Trautwein & Morin, 2009; Muthén, 2003), the

statistical adequacy of the solution (e.g., absence of negative variance estimates; Bauer &

Curran, 2004), and various other statistical indicators (McLachlan & Peel, 2000). Among

the statistical indicators considered are the Akaike Information Criterion (AIC), the

Bayesian information criterion (BIC), the Consistent AIC (CAIC), the sample-size

adjusted BIC (ABIC), the Lo, Mendell, and Rubin (2001) likelihood ratio test (LMR), the

Bootstrap Likelihood Ratio Test (BLRT), and the entropy. Lower values on AIC, CAIC,

BIC, and ABIC suggest a better-fitting model. Both the LMR and BLRT compare a k-

profile model with a (k-1)-profile model. A statistically significant p value indicates that

the (k-1) profile model should be rejected in favor of a less parsimonious k-profile model.

However, because these tests are based on statistical significance assumptions, the class

enumeration procedure that follows these indices can still be influenced by sample size

(Marsh et al., 2009) and may continue to favor the addition of profiles. Consequently,

these indices can be graphically presented through “elbow plots” illustrating the gains

associated with additional profiles (Morin, et al., 2011). In these plots, the point after

which the slope flattens out indicates the optimal number of profiles. Elbow plots for all

profile analyses in the 6 domains are presented in Figure S2. Finally, the entropy

indicates the precision with which participants are classified in the profiles (varying from

0 to 1; highest values being better) and should be considered with other indices to

determine the optimal number of profiles (Lubke & Muthén, 2007).

To test whether there are differences among retained profiles on measures of self-

efficacy at Time 1 and Time 2 and on college program attendance, the auxiliary function


in Mplus was used. This model command offers the possibility to contrast profiles to

determine if they differ on a dependent variable. Table S6 presents these Cohen’s d

comparisons between profiles for motivational types as well as indicators of vocational

adjustment within each domain.

Results

Missing Data Analysis

As in most longitudinal datasets, many participants (66%) did not complete the T2

questionnaire, which assessed self-efficacy toward each activity in each RIASEC domain.

Consequently, participants who did not complete the T2 questionnaire were compared on

Time 1 to those who completed both time measurements with invariance analyses. As

presented in Table S7, results showed no differences on the various model parameters

(Models 1 to 5) across RIASEC domains. Moreover, none of the fit indices improved

substantially when means were free to vary across groups, indicating that there were no

substantial latent mean differences in latent constructs between participants with

complete and partial data. This led us to conclude that no systematic patterns of

missingness (e.g., MNAR) could seriously bias subsequent results. To account for

missing data in the SEM analyses, full information maximum likelihood (FIML)

estimation was used to compute the product of individual likelihood functions in order to

estimate the analysis parameters for the whole sample. Using a FIML procedure to handle

missing data is considered superior to using listwise deletion and other ad hoc methods

such as mean substitution (Enders, 2010; Davey, Shanahan, & Schafer, 2001; Graham,

2009; Peugh & Enders, 2004).

Results From Additional Models


To verify the adequacy of measurement models, six CFA models were estimated

(one per domain of interests; Models S1) in which the 6 latent factors representing

motivation types and self-efficacy at both waves were correlated. Fit indices for these

various models were all acceptable (Table S2), with CFI and TLI values over .95 and

RMSEA, SRMR values below .06. Correlations among latent factors representing

motivational types within each RIASEC dimension are presented in Table S1. Results

supported the presence of a motivation continuum within each RIASEC domain. Most

correlations follow the simplex-like pattern theorized by SDT, which specifies that

correlations of more proximal motivation types on the continuum should be higher than

correlations of more distal pairs (Howard, Gagné, & Bureau, 2017). However, it should

be noted that, for E and C dimensions, the simplex-like pattern was less supported.

Results from all additional models predicting self-efficacy in every domain (Models S2;

see Tables S2 and S3) and college program domain (Models 3; see Tables S2 and S4) are

presented in the tables. However, in models presented in the main manuscript, self-

efficacy at T1 is used as a control variable, which could have led to suppression effect.

Moreover, regression coefficients in non-corresponding domains were not included in the

manuscript. For these reasons, Tables S4 and S5 present results based on SEM analyses

without T1 self-efficacy as a covariate. Moreover, results in these two tables show

regression coefficients in non-corresponding domains.

Latent Profiles Analyses

For each RIASEC domain of activities, LPA was performed to identify

intraindividual patterns of motivation at T1. Every LPA model was tested with

constrained and unconstrained variances across the different profiles. Solutions for the


constrained variance model were the most appropriate and provided the most

theoretically sound results. Consequently, only results from the constrained variances

solutions were reported in the article. Based on elbow plots (see Figure S2), fit indices

(see Table S5), theoretical considerations, and the number of participants in each profile,

a three-profile or a four-profile solution were deemed optimal for each RIASEC

dimension. Notably, the log-likelihood was replicated at least 50 times for each of the

selected solutions, providing support for their robustness. Solutions with more than three

or four profiles were not retained because they contained very few participants in some

profiles (< 10%) which which can threaten the reliability of subsequent comparisons.

More specifically, the fourth profile for the Realistic domain included 8% of the sample,

while it was 6% for Investigative, 8% for Artistic, 7% for Enterprising, and 5% for

Conventional. Only the fourth profile on the Social dimension included 15% of the

sample. Inspection of this profile revealed the same motivational pattern as those in some

3-profile solutions: an introjected one. A 4-profile solution was thus favored for the

Social dimension. Table S6 present Cohen d for comparing motivational levels across

profiles in a specific domain.


References

Bauer, D. J., & Curran, P. J. (2004). The integration of continuous and discrete latent

variable models: Potential problems and promising opportunities. Psychological

methods, 9, 3-29. dx.doi.org/10.1037/1082-989X.9.1.3

Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K.A.

Bollen & J. Scott Long (Eds.), Testing Structural Equation Models (pp. 136-154).

London, UK: Sage focus editions.

Byrne, B. M. (2013). Structural equation modeling with Mplus: Basic concepts,

applications, and programming. New York, NY: Routledge.

Davey, A., Shanahan, M. J., & Schafer, J. L. (2001). Correcting for selective nonresponse

in the National Longitudinal Survey of Youth using multiple imputation. Journal

of Human Resources, 36, 500-519. DOI: 10.2307/3069628

Enders, C.K. (2010). Applied missing data analysis. New York: Guilford.

Graham, J.W. (2009). Missing data analysis: Making it work in the real world. Annual

Review of Psychology, 60, 549-576.

Doi.org/10.1146/annurev.psych.58.110405.085530.

Hipp, J. R., & Bauer, D. J. (2006). Local solutions in the estimation of growth mixture

models. Psychological Methods, 11, 36-53. dx.doi.org/10.1037/1082-

989X.11.1.36

Hogan, T. P. (2007). Psychological testing: a practical introduction (2nd ed.). Hoboken,

NJ: John Wiley.

Holland, J. L. (1991). L’orientation par soi-même : édition canadienne-française.

Odessa, FL: Psychological Assessment Resources.

Lubke, G., & Muthén, B. O. (2007). Performance of factor mixture models as a function

of model size, covariate effects, and class-specific parameters. Structural

Equation Modeling, 14, 26-47.





Marsh, H. W., Lüdtke, O., Trautwein, U., & Morin, A. J. (2009). Classical latent profile

analysis of academic self-concept dimensions: Synergy of person-and variable-

centered approaches to theoretical models of self-concept. Structural Equation

Modeling, 16, 191-225. doi.org/10.1080/10705510902751010

McDonald, R. P. (1970). The theoretical foundations of principal factor analysis,

canonical factor analysis, and alpha factor analysis. British Journal of

Mathematical and Statistical Psychology, 23, 1-21. doi.org/10.1111/j.2044-

8317.1970.tb00432.x

McLachlan, G., & Peel, D. (2000). Multivariate normal mixtures. Finite Mixture Models,

New York : Wiley, 81-116.

Morin, A. J., Maiano, C., Nagengast, B., Marsh, H. W., Morizot, J., & Janosz, M. (2011).

General growth mixture analysis of adolescents' developmental trajectories of

anxiety: The impact of untested invariance assumptions on substantive

interpretations. Structural Equation Modeling: A Multidisciplinary Journal, 18,

613-648. doi.org/10.1080/10705511.2011.607714

Muthén, B. O. (2002). Beyond SEM: General latent variable modeling.

Behaviormetrika, 29, 81-117. doi.org/10.2333/bhmk.29.81

Muthén, B. (2003). Statistical and substantive checking in growth mixture modeling:

comment on Bauer and Curran. Psychological Methods, 8, 369-77; discussion

384-93. doi.org/10.1037/1082-989X.8.3.369

Muthén, L. K., & Muthén, B. O. (2017). Mplus user’s guide. Eighth Edition. (Eighth

Ed.). Los Angeles, CA: Muthén & Muthén.

Peugh, J. L., & Enders, C. K. (2004). Missing data in educational research: A review of

reporting practices and suggestions for improvement. Review of educational

research, 74, 525-556. doi.org/10.3102/00346543074004525

https://doi.org/10.3102%2F00346543074004525


https://www.ncbi.nlm.nih.gov/pubmed/14596497

https://doi.org/10.2333/bhmk.29.81

https://doi.org/10.1080/10705511.2011.607714



https://doi.org/10.1080/10705510902751010


Peugh, J., & Fan, X. (2013). Modeling unobserved heterogeneity using latent profile

analysis: A Monte Carlo simulation. Structural Equation Modeling: A

Multidisciplinary Journal, 20, 616-639. doi.org/10.1080/10705511.2013.824780

Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common

method biases in behavioral research: A critical review of the literature and

recommended remedies. Journal of applied psychology, 88, 879-903.

dx.doi.org/10.1037/0021-9010.88.5.879

Schumacker, E. R., & Lomax, G. R. (1996). A beginner’s guide to structural equation

modeling. Mahwah, NJ: Lawrence Erlbaum.

Yu, C. Y. (2002). Evaluation of model fit indices for latent variable models with

categorical and continuous outcomes. Paper presented at the Annual Conference

of the American Educational Research Association, April 2002, New Orleans.

http://psycnet.apa.org/doi/10.1037/0021-9010.88.5.879

https://doi.org/10.1080/10705511.2013.824780


Table S1.

Correlations among Latent Factors for Intrinsic, Identified, Introjected, and External Regulation for Each RIASEC DomainMean SD Intrinsic Identified Introjected External SE T1 SE t2

RealisticIntrinsic 2.15 1.29 -Identified 2.26 1.23 .71 -Introjected 1.52 .96 .24 .32 -External 2.68 1.34 .11 .19 .26 -SE T1 1.94 .99 .76 .64 .29 .18 -SE T2 2.09 1.01 .69 .54 .16 .12 .77 -InvestigativeIntrinsic 2.09 1.16 -Identified 2.45 1.24 .76 -Introjected 1.60 .96 .24 .28 -External 2.87 1.25 .02 .16 .18 -SE T1 2.60 1.15 .72 .73 .19 .19 -SE T2 2.90 1.10 .61 .61 .09 .18 .82 -Program .27 .45 .48 .52 .10 .09 .55 .50ArtisticIntrinsic 2.66 1.24 -Identified 2.44 1.17 .81 -Introjected 1.58 .93 .14 .23 -External 2.75 1.20 .05 .10 .20 -SE T1 2.58 1.03 .77 .73 .16 .12 -SE T2 2.76 .99 .74 .66 .14 .05 .79Program .10 .30 .25 .32 .02 .10 .35 .33SocialIntrinsic 3.10 1.18 -Identified 3.26 1.12 .86 -Introjected 1.82 1.13 .18 .22 -External 2.87 1.18 .06 .17 .36 -SE T1 3.19 .90 .79 .76 .19 .14 -SE T2 3.47 .91 .65 .68 .11 .05 .81Program .62 .49 .31 .25 .08 -.05 .35 .37EnterprisingIntrinsic 2.54 1.21 -Identified 2.95 1.21 .80 -Introjected 1.74 1.07 .28 .29 -External 3.03 1.16 .13 .27 .27 -SE T1 2.94 .98 .72 .72 .22 .24 -SE T2 3.22 1.00 .54 .57 .14 .13 .71ConventionalIntrinsic 2.17 1.18 -Identified 2.85 1.23 .76 -Introjected 1.65 1.02 .28 .26 -External 3.01 1.21 .14 .28 .21 -SE T1 2.82 1.15 .64 .66 .19 .24 -SE T2 3.15 1.13 .47 .48 .06 .24 .73

Note. SD = standard deviation; SE = self-efficacy; T1 = Time 1; T2 = Time 2.

Running Head : RIASEC interests type 36

Table S2Fit Indices for All CFA and SEM Models Presented in the Manuscript and Supplementary Materials

CFI TLI X2 df RMSEA SRMR WRMRRealistic

1r. SEM with SE at T2 controlling for SE at T1 .966 .952 598.78 193 .046 .034S1r. CFA .972 .957 509.77 177 .044 .032S2r. SEM with SE at T2 in all domains .948 .939 1471.50 671 .035 .051S3r. SEM with all programs at T1 .938 .904 212.53 110 .031 - .529

Investigative1i. SEM with SE at T2 controlling for SE at T1 .988 .983 320.14 193 .026 .0262i. SEM with corresponding program at T1

controlling for SE at T1.992 .987 215.50 135 .025 .022

S1i. CFA .992 .987 269.00 177 .023 .024S2i. SEM with SE at T2 in all domains .961 .954 1201.09 671 .028 .051S3i. SEM with all programs at T1 .972 .957 166.38 110 .023 - .496

Artistic1. SEM with SE at T2 controlling for SE at T1 .953 .933 658.91 193 .050 .0572. SEM with corresponding program at T1


S1a. CFA .982 .972 354.57 177 .032 .036S2a. SEM with SE at T2 in all domains .929 .918 1543.27 671 .037 .060 -S3a. SEM with all program at T1 .909 .858 245.46 110 .036 - .787

Social 1s. SEM with SE at T2 controlling for SE at T1 .977 .967 403.39 193 .033 .0362s. SEM with corresponding program at T1


S1s. CFA .985 .977 311.43 177 .028 .030S2s. SEM with SE at T2 in all domainsS3s. SEM with all program at T1

Entreprising1e. SEM with SE at T2 controlling for SE at T1 .993 .990 268.92 193 .020 .027S1e. CFA .997 .996 205.22 177 .013 .025S2e. SEM with SE at T2 in all domains .951 .944 1244.19 671 .030 .054 -S3e. SEM with all programs at T1 .970 .954 157.79 110 .021 - .583

Conventionnel 1c. SEM with SE at T2 controlling for SE at T1 .989 .985 322.33 193 .026 .023S1c. CFA .991 .986 289.15 177 .025 .022S2c. SEM with SE at T2 in all domains .963 .957 1224.89 671 .029 .051 -S3c. SEM with all programs at T1 .966 .947 174.02 110 .024 - .471

Note. CFI = Comparative Fit Index; TLI = Tucker-Lewis Fit Index; RMSEA =: Root Mean Square Error of Approximation; df = degrees of freedom; SRMR = Standardized Root Mean Square Residual; SE = self-efficacy; T1 = Time 1; T2 = Time 2.

Running Head : RIASEC interests type 37

Table S3 Means for Target Loadings, Standard Deviations, Range and Regression Coefficients for Additional Models Predicting Self-Efficacy in Each RIASEC Domain from Motivation in One Domain (Models S2 in Table S2)

Model Type of Interest

Mean loading

Mean SD

loading

Loadings range

predict SE in R

predict. SE in I

predict. SE A

predict. SE in S

predict. SE in E

predict SE in C

RealisticIntrinsic 0.85 0.06 0.77-.91 .67* .29* -.20* -.41* .01 .19*

Model S2r Identified 0.86 0.06 0.78-.92 .08 -.11 .09 .11 .07 -.06Introjected 0.88 0.05 0.83-.93 -.06 -.16* -.06 .05 -.07 -.17*External 0.85 0.05 0.79-.90 .02 .12* .06 .08 .08 .15*InvestigativeIntrinsic 0.82 0.05 0.75-.87 .31* .43* -.10 -.24* -.18* .06

Model S2i Identified 0.82 0.04 0.79-.86 .02 .31* .01 .06 .26* .30*Introjected 0.85 0.03 0.81-.88 -.10* -.10 -.05 .05 -.01 -.13*External 0.80 0.05 0.76-.85 -.06 .09 -.05 .02 .12 .20*ArtisticIntrinsic 0.71 0.17 0.53-.89 -.16 .01 .76* .15 -.10 .02

Model S2a Identified 0.74 0.14 0.58-.87 .07 -.06 .09 .11 .12 -.10Introjected 0.82 0.08 0.73-.90 -.06 -.05 -.02 .06 -.03 -.06External 0.70 0.14 0.56-.84 .09 .01 -.01 -.06 .03 .11SocialIntrinsic 0.71 0.09 0.61-.82 -.12 .11 -.01 .26 -.16 .03

Model S2s Identified 0.70 0.09 0.60-.82 -.13 -.28 .26 .51* .05 -.21Introjected 0.84 0.05 0.79-.90 -.10 -.11 -.18* -.00 -.10 -.14*External 0.71 0.08 0.62-.82 .11 .09 .02 -.10 .10 .16*EnterprisingIntrinsic 0.80 0.06 0.73-.85 .05 -.01 -.31* -.23* .28* .25*

Model S2e Identified 0.80 0.06 0.74-.86 .08 .06 .22 .16 .39* .16Introjected 0.86 0.04 0.83-.91 -.11 -.09 -.11 .01 -.06 -.08External 0.79 0.07 0.69-.85 .01 .11 -.04 -.03 -.02 .14*ConventionalIntrinsic 0.86 0.03 0.83-.89 .10 .10 -.33* -.29* .11 .33*

Model S2c Identified 0.83 0.04 0.78-.86 .10 .15 .14 .08 .28* .25*Introjected 0.88 0.02 0.86-.92 -.14* -.08 -.13* -.01 -.04 -.10*External 0.82 0.04 0.77-.85 -.08 .06 .05 .02 .09 .15*

Note. This table describes models with four latent factors for types of motivation predicting self-efficacy in all RIASEC domain with standardized regression coefficients being presented. SE = Standard Error; R = Realistic, I = Investigative, A = Artistic, S = Social, E = Enterprising, C = Conventional; SE = Self-efficacy; SD = Standard Deviation. *p < .05.

Self-determination theory and vocational interests 38

Table S4 Means for Target Factor Loadings, Standard Deviations, Ranges, and Regression Coefficients for the Prediction of College Program Attendance (Models S3 in Table S2)

Model Type of program Mean target

loading

SD target

loading

Range target

loading

predict. program

investigative

predict. program artistic

predict. program.

socialRealistic

Model S3r Intrinsic 0.85 0.06 0.77-.91 .17* -.26* -.05Identified 0.86 0.06 0.78-.91 .07 .20 -.14Introjected 0.87 0.06 0.82-.94 -.12* -.18* .18*External 0.84 0.09 0.76-.92 .10 .04 -.10Investigative

Model S3i Intrinsic 0.82 0.05 0.77-.88 .22* -.38* -.11Identified 0.82 0.02 0.80-.85 .48* .15 -.45*Introjected 0.85 0.04 0.82-.90 -.06 -.11 .10*External 0.80 0.04 0.74-.84 .05 -.06 -.02Artistic

Model S3a Intrinsic 0.69 0.10 0.55-.78 -.09 .10 .10Identified 0.74 0.11 0.58-.83 -.17 .46* -.16Introjected 0.82 0.10 0.74-.91 .03 -.14* .05External 0.69 0.07 0.61-.75 .08 .14* -.13*Social

Model S3s Intrinsic 0.72 0.08 0.64-83 -.28* -.16 .36*Identified 0.70 0.11 0.59-.85 -.15 .15 .03Introjected 0.84 0.04 0.79-.89 -.00 -.15* .08External 0.71 0.06 0.65-.80 .11 .10 -.14*Enterprising

Model S3e Intrinsic 0.80 0.03 0.75-.83 -.04 -.52* .29*Identified 0.81 0.04 0.75-.84 .05 .52* -.29*Introjected 0.86 0.06 0.79-.93 -.07 -.16* .12*External 0.79 0.08 0.68-.86 .21* -.02 -.17*Conventional

Model S3c Intrinsic 0.87 0.03 0.83-.89 .11 -.29* -.01Identified 0.83 0.05 0.78-.88 .18* .12 -.17*Introjected 0.88 0.04 0.83-.93 -.03 -.13 .07External 0.82 0.03 0.78-.85 .11 -.03 -.08

Note. SD = Standard Deviation. *p < .05


Table S5. Results from the Latent Profile Analyses

Model LL #fp Scaling AIC CAIC BIC ABIC Entropy aLMR BLRTRealistic

1 profile -5445.05 8 1.02 10906.11 10953.16 10945.16 10919.75 na na na2 profile -4952.10 13 1.23 9930.20 10006.66 9993.66 9952.37 0.90 ≤ 0.001 ≤ 0.0013 profile -4699.47 18 1.48 9434.94 9540.81 9522.81 9465.64 0.89 ≤ 0.001 ≤ 0.0014 profile -4485.02 23 1.54 9016.05 9151.32 9128.32 9055.27 0.91 .002 ≤ 0.0015 profile -4331.20 28 1.90 8718.39 8883.07 8855.07 8766.15 0.92 .244 ≤ 0.0016 profile -4194.87 33 1.66 8455.74 8649.83 8616.83 8512.02 0.93 .028 ≤ 0.0017 profile -4096.85 38 1.62 8269.69 8493.18 8455.18 8334.50 0.91 ≤ 0.001 ≤ 0.0018 profile -3998.67 43 1.98 8083.34 8336.24 8293.24 8156.67 0.92 .102 ≤ 0.001

Investigative1 profile -5782.35 8 0.98 11580.70 11627.75 11619.75 11594.35 na na na2 profile -5251.86 13 1.22 10529.72 10606.18 10593.18 10551.89 0.88 ≤ 0.001 ≤ 0.0013 profile -5074.60 18 1.35 10185.20 10291.06 10273.06 10215.90 0.88 ≤ 0.001 ≤ 0.0014 profile -4887.28 23 1.51 9820.57 9955.84 9932.84 9859.79 0.93 .009 ≤ 0.0015 profile -4744.56 28 1.50 9545.12 9709.80 9681.80 9592.87 0.94 .003 ≤ 0.0016 profile -4647.77 33 1.54 9361.54 9555.63 9522.63 9417.82 0.94 .024 ≤ 0.0017 profile -4566.15 38 1.54 9208.31 9431.80 9393.80 9273.11 0.95 .039 ≤ 0.0018 profile -4507.95 43 1.51 9101.91 9354.81 9311.81 9175.24 0.90 ≤ 0.001 ≤ 0.001

Artistic1 profile -5987.59 8 0.96 11991.19 12038.24 12030.24 12004.83 na na na2 profile -5418.39 13 1.13 10862.79 10939.25 10926.25 10884.96 0.89 ≤ 0.001 ≤ 0.0013 profile -5213.07 18 1.54 10462.13 10568.00 10550.00 10492.83 0.91 .047 ≤ 0.0014 profile -5025.73 23 1.47 10097.46 10232.73 10209.73 10136.68 0.92 .002 ≤ 0.0015 profile -4889.36 28 1.54 9834.73 9999.41 9971.41 9882.48 0.90 .050 ≤ 0.0016 profile -4788.46 33 1.45 9642.93 9837.01 9804.01 9699.21 0.91 .009 ≤ 0.0017 profile -4688.50 38 1.62 9453.00 9676.49 9638.49 9517.81 0.92 .182 ≤ 0.0018 profile -4608.96 43 1.73 9303.92 9556.82 9513.82 9377.25 0.93 .425 ≤ 0.001

Social1 profile -4716.12 8 0.82 9448.23 9495.28 9487.28 9461.87 na na na2 profile -4200.08 13 1.01 8426.16 8502.62 8489.62 8448.33 0.83 ≤ 0.001 ≤ 0.0013 profile -3916.72 18 1.06 7869.44 7975.31 7957.31 7900.14 0.87 ≤ 0.001 ≤ 0.0014 profile -3700.04 23 1.22 7446.08 7581.35 7558.35 7485.30 0.88 ≤ 0.001 ≤ 0.0015 profile -3552.35 28 1.33 7160.70 7325.38 7297.38 7208.46 0.89 .018 ≤ 0.0016 profile -3470.56 33 1.71 7007.12 7201.21 7168.21 7063.40 0.89 .463 ≤ 0.0017 profile -3402.49 38 1.38 6880.97 7104.46 7066.46 6945.78 0.80 .062 ≤ 0.0018 profile -3339.58 43 1.37 6765.16 7018.06 6975.06 6838.49 0.88 .006 ≤ 0.001

Enterprising1 profile -5878.88 8 0.87 11773.77 11820.82 11812.82 11787.41 na na na2 profile -5407.99 13 1.07 10841.97 10918.43 10905.43 10864.14 0.81 ≤ 0.001 ≤ 0.0013 profile -5184.68 18 1.22 10405.35 10511.22 10493.22 10436.05 0.83 ≤ 0.001 ≤ 0.0014 profile -4975.85 23 1.21 9997.69 10132.97 10109.97 10036.92 0.90 ≤ 0.001 ≤ 0.0015 profile -4846.17 28 1.23 9748.34 9913.02 9885.02 9796.10 0.87 ≤ 0.001 ≤ 0.0016 profile -4767.37 33 1.72 9600.74 9794.83 9761.83 9657.02 0.88 .501 ≤ 0.0017 profile -4684.32 38 1.47 9444.63 9668.13 9630.13 9509.44 0.89 .050 ≤ 0.0018 profile -4611.79 43 1.38 9309.58 9562.48 9519.48 9382.91 0.91 .025 ≤ 0.001

Conventional1 profile -5803.65 8 0.95 11623.30 11670.35 11662.35 11636.94 na na na2 profile -5342.17 13 1.13 10710.34 10786.80 10773.80 10732.51 0.85 ≤ 0.001 ≤ 0.0013 profile -5138.81 18 1.32 10313.61 10419.48 10401.48 10344.31 0.88 .001 ≤ 0.0014 profile -4951.25 23 1.37 9948.50 10083.77 10060.77 9987.73 0.91 .002 ≤ 0.0015 profile -4808.38 28 1.51 9672.77 9837.45 9809.45 9720.52 0.92 .014 ≤ 0.0016 profile -4698.75 33 1.42 9463.50 9657.59 9624.59 9519.78 0.93 .007 ≤ 0.0017 profile -4610.74 38 2.51 9297.48 9520.97 9482.97 9362.28 0.93 .783 ≤ 0.0018 profile -4528.23 43 1.57 9142.47 9395.37 9352.37 9215.80 0.94 .047 ≤ 0.001

Note. LL= loglikelihood; #fp = number of free parameters; AIC = Akaïke Information Criteria; CAIC = Constant AIC; BIC = Bayesian Information Criteria; ABIC = Sample-Size adjusted BIC; aLMR = Lo-Mendell-Rubin adjusted LRT; BLRT = Bootstrapped Likelihood Ratio Test; na not applicable.


Table S6Cohen’s d Coefficient for Effect Sizes of Between-Profile Mean Comparisons

Self-determined vs introjected

Self-determined vs moderated

Self-determined vs low

Introjected vs moderated

Introjected vs low

Moderated vs low

Realistic Intrinsic -0.88 - -2.84 - -1.96 - Identified -0.24 - -2.18 - -1.97 - Introjected 4.89 - -0.14 - -5.03 - External regulation

0.63 - -0.19 - -0.82 -

Self-efficacy T1 -0.16 - -1.70 - -1.65 - Self-efficacy T2 -0.54 - -1.18 - -0.56 - Adjusted mean at T2

-0.31 - -0.42 - -0.10 -

Investigative Intrinsic - -3.41 -6.40 - - -2.98 Identified - -1.60 -3.86 - - -2.26 Introjected - -0.05 -0.64 - - -0.59 External regulation

- 0.34 -0.01 - - -0.35

Self-efficacy T1 - -1.02 -2.38 - - -1.41 Self-efficacy T2 - -0.55 -1.04 - - -0.51 Adjusted mean at T2

- -0.13 -0.19 - - -0.08

Investigative program

- -0.36 -1.42 - - -0.74

Artistic Intrinsic -1.29 - -3.21 - -1.92 - Identified -0.66 - -2.62 - -1.96 - Introjected 4.64 - 0.22 - -4.42 - External regulation

0.59 - 0.04 - -0.56 -


-0.13 - -0.33 - -0.20 -

Artistic program -0.18 - -0.35 - -0.15 -Social Intrinsic -0.58 -0.42 -5.44 -2.22 -4.86 -2.64 Identified -0.32 -2.73 -5.32 -2.41 -5.00 -2.59 Introjected 3.97 0.72 0.35 -3.25 -3.62 -0.37 External regulation

1.28 0.28 0.54 -1.00 -1.20 -0.20

Self-efficacy T1 -0.39 -1.67 -2.95 -1.23 -2.40 -1.32 Self-efficacy T2 -0.26 -0.75 -1.37 -0.41 -0.91 -0.59 Adjusted mean at T2

-0.15 -0.20 -0.42 -0.06 -0.29 -0.25

Social program -0.08 -0.35 -0.68 -0.26 -0.58 -0.30Enterprising Intrinsic - -2.76 -5.07 - - -2.31 Identified - -2.33 -4.95 - - -2.62 Introjected - -0.46 -0.85 - - -0.39 External regulation

- -0.22 -0.58 - - -0.36

Self-efficacy T1 - -1.28 -2.31 - - -1.17 Self-efficacy T2 - -0.47 -0.97 - - -0.43 Adjusted mean at T2

- -0.14 -0.27 - - -0.17

Conventional Intrinsic -0.75 - -2.73 - -1.98 - Identified -0.65 - -2.38 - -1.73 - Introjected 4.37 - 0.11 - -4.26 - External regulation

0.36 - -0.28 - -0.64 -


-0.13 - -0.03 - -0.11 -

Note. Cohen’s d value around ±.20 are considered small, around ±.50 are considered medium, and around ±.80 or higher are considered large.


Table S7Fit Indices for CFA Models Testing the Equivalence of Participants with Partial and Complete Data

Domain Npar S-B 2 df CFI TLI RMSEA [90%CI] SRMR BIC AICRealistic

1-Loading.+ intercepts 132 466.52 172 .962 .948 .059 [.053-.066] .033 40911.55 40267.202-Loading.+ intercepts +u 116 457.75 188 .966 .956 .054 [.048-.061] .034 40852.31 40286.063-Loading.+ intercepts +u + corr u 92 482.48 212 .966 .961 .051 [.045-.057] .034 40740.93 40291.844-Loading.+ intercepts +u + corr u + var/cov 78 547.12 226 .959 .957 .054 [.048-.060] .074 40734.53 40353.785- Model 4 + means fixed at 0 74 552.35 230 .959 .957 .054 [.048-.059] .074 40710.36 40349.14

Investigative1-Loading.+ intercepts 132 229.60 172 .992 .988 .026 [.016-.035] .027 42704.67 42060.322-Loading.+ intercepts +u 116 240.25 188 .993 .990 .024 [.013-.032] .029 42628.31 42062.063-Loading.+ intercepts +u + corr u 92 283.36 212 .990 .988 .026 [.017-.034] .030 42524.93 42075.844-Loading.+ intercepts +u + corr u + var/cov 78 384.74 226 .977 .976 .038 [.031-.044] .074 42559.94 42179.195- Model 4 + means fixed at 0 74 392.04 230 .977 .976 .038 [.032-.044] .074 42540.74 42179.52

Artistic1-Loading.+ intercepts 132 344.42 172 .973 .962 .045 [.038-.052] .042 46290.97 45646.632-Loading.+ intercepts +u 116 365.43 188 .972 .964 .044 [.037-.051] .043 46226.22 45659.983-Loading.+ intercepts +u + corr u 92 397.43 212 .970 .967 .042 [.036-.049] .043 46114.99 45665.904-Loading.+ intercepts +u + corr u + var/cov 78 537.01 226 .950 .947 .053 [.047-.059] .083 46187.05 45806.305- Model 4 + means fixed at 0 74 538.45 230 .951 .949 .052 [.047-.058] .083 46160.63 45799.41

Social1-Loading.+ intercepts 132 295.41 172 .980 .972 .038 [.031-.046] .040 47341.17 46696.832-Loading.+ intercepts +u 116 300.43 188 .982 .977 .035 [.027-.042] .040 47246.93 46680.693-Loading.+ intercepts +u + corr u 92 326.48 212 .982 .979 .033 [.026-.040] .040 47123.19 46674.104-Loading.+ intercepts +u + corr u + var/cov 78 362.79 226 .978 .977 .035 [.028-.042] .062 47066.88 46686.135- Model 4 + means fixed at 0 74 366.74 230 .978 .977 .035 [.028-.042] .063 47042.58 46681.35

Enterprising1-Loading.+ intercepts 132 193.08 172 .997 .996 .016 [.000-.027] .028 43803.74 43159.402-Loading.+ intercepts +u 116 208.90 188 .997 .996 .015 [.000-.026] .029 43727.53 43161.283-Loading.+ intercepts +u + corr u 92 234.20 212 .997 .996 .015 [.000-.025] .029 43602.84 43153.754-Loading.+ intercepts +u + corr u + var/cov 78 303.70 226 .989 .988 .027 [.018-.034] .073 43590.40 43209.655-Model 4 + means fixed at 0 74 304.99 230 .989 .989 .026 [.017-.033] .073 43563.81 43202.59

Conventional1-Loading.+ intercepts 132 279.39 172 .987 .981 .036 [.028-.043] .028 40912.18 40267.832-Loading.+ intercepts +u 116 279.31 188 .989 .986 .032 [.023-.039] .028 40833.14 40266.90


3-Loading.+ intercepts +u + corr u 92 299.60 212 .989 .988 .029 [.021-.036] .027 40696.81 40247.724-Loading.+ intercepts +u + corr u + var/cov 78 361.58 226 .983 .982 .035 [.028-.042] .072 40675.87 40295.125- Model 4 + means fixed at 0 74 366.33 230 .983 .982 .035 [.028-.041] .073 40653.17 40291.95

Note. Npar = Number of parameters in the model; CFI = Comparative Fit Index; TLI = Tucker-Lewis Fit Index; CI = Confidence Interval; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardized Root Mean Square Residual; BIC = Bayesian Information Criteria; AIC = Akaike Information Criteria.


Figure S1. The SEM tested in each RIASEC domain. Six models were tested. Cross-sectional correlations among error terms of parallel items were estimated to avoid inflated correlations among exogeneous latent factors. Longitudinal correlations among error terms of parallel items for self-efficacy were also estimated. When program attendance was used as an endogenous variable (instead of T2 self-efficacy), no longitudinal correlated uniquenesses were estimated.


7000

7500

8000

8500

9000

9500

10000

10500

11000

11500

1 2 3 4 5 6 7 8n profiles

Realistic

AIC CAIC BIC ABIC

8500

9000

9500

10000

10500

11000

11500

12000

1 2 3 4 5 6 7 8n profiles

Investigative

AIC CAIC BIC ABIC

8500

9000

9500

10000

10500

11000

11500

12000

12500

1 2 3 4 5 6 7 8n profiles

Artistic

AIC CAIC BIC ABIC

6000

6500

7000

7500

8000

8500

9000

9500

10000

1 2 3 4 5 6 7 8n profiles

Social

AIC CAIC BIC ABIC

8500

9000

9500

10000

10500

11000

11500

12000

12500

1 2 3 4 5 6 7 8n profiles

Enterprising

AIC CAIC BIC ABIC

8500

9000

9500

10000

10500

11000

11500

12000

1 2 3 4 5 6 7 8n profiles

Conventional

AIC CAIC BIC ABIC

Figure S2. Elbow Plots of the Akaike Information Criteria for all RIASEC models

Date post:	05-Sep-2019
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times

corpus.ulaval.ca€¦ · Web viewT1 introjected and external regulations only yielded three...

Documents