Is the Effect of Self-Efficacy on Job/Task Performance an Epiphenomenon?

1
Is the Effect of Self-Efficacy on Job/Task
Performance an Epiphenomenon?

• Timothy A. Judge
• Christine Jackson
• John C. Shaw
• Brent A. Scott
• Bruce Louis Rich
• University of Florida

2
Self-Efficacy

• Described as the theory heard round the world
• Albert Bandura deemed third most influential
  psychologist in history
• Self-efficacy has been the subject of 8,944
  studies
• Has been applied to health, child development,
  sports, clinical psychology, education and, bien
  sûr, I-O psychology

3
Self-Efficacy in I-O/OB

• Learning in training (Martocchio, 1994)
• Naval performance and seasickness (Eden Zuk,
  1995)
• Volunteering for reenlistment (Eden Kinnar,
  1991)
• Speed of re-employment (Eden Aviram, 1993)
• Sales performance (Barling Beattie, 1983)
• Managerial performance (Wood et al., 1990)
• Academic performance (Wood Locke, 1987)
• Reaction to stressors (Jex Bliese, 1999)
• Success of collegiate hockey teams (Feltz
  Lirgg, 1998)
• Salary negotiation (Stevens Gist, 1997)
• Participation in union activities (Bulger
  Mellor, 1997)
• Newcomer socialization and adjustment (Saks,
  1995)
• Creativity (Redmond, Mumford, Teach, 1993)
• Coping with career-related events (Stumpf et al.,
  1987)
• Skill acquisition (Mitchell et al., 1994)
• Adaptation to advanced technology (Hill et al.,
  1987)

4
What Is Validity of Self-Efficacy?

• Stajkovic and Luthans (1998) meta-analyzed the
  self-efficacy performance relationship
• .34
• few cognitive determinants of behavior... have
  received as ample and consistent empirical
  support as the concept of self-efficacy (p. 240)
• However, with some exceptions (Chen, Casper,
  Cortina, 2001 Phillips Gully, 1997), most
  estimates do not take distal controls into account

5
Role of Individual Differences

• Self-efficacy is related to various individual
  differences
• Intelligence (Phillips Gully, 1997)
• Personality (Judge Ilies, 2002)
• Experience (Shea Howell, 2000)
• These individual differences are thought to be
  more distal than self-efficacy, and thus less
  direct
• But, this has not been tested in a path model

6
Conceptual Model
Self-set goals
Job/Task Performance
Self-efficacy
General mental ability
Conscien-tiousness
Emotional stability
Extraversion
Experience
7
MethodLiterature Search

• In forming the correlation matrix that was used
  as input into the LISREL model, we took two steps
• where meta-analytic estimates were available, we
  used these directly
• Barrick, Mount, Judge, 2001 Judge Bono,
  2001 Judge Ilies, 2002 Ones, Viswesvaran,
  Reiss, 1996 Quiñones, Ford, Teachout, 1995
  Schmidt Hunter, 1998 Stajkovic Luthans, 1998
• where meta-analytic estimates were unavailable
  (involving GMA and experience), we performed our
  own meta-analyses

8
MethodSearch Results
9
MethodModerator Coding

• Several moderator variables were coded in the
  present study
• Type of measure Likert scale or grid
  (self-efficacy strength and magnitude)
• Job/task complexity
• Knowledge of results
• Pending further analysis
• Feedback, performance measure, study setting,
  sample
• Hierarchical moderator analyses (?)

10
Meta-Analysis Procedures

• Used procedures developed by Hunter and Schmidt
  (1990)
• We corrected each primary correlation for
  attenuation due to unreliability, and then
  computed the sample-weighted average corrected
  correlation
• For studies that did not report reliabilities, we
  used the mean of the reliabilities reported for
  the variables of interest

11
Overall Correlation Matrix
12
Results Full Mediation Model
Self-set goals
.16
Job/Task Performance
.50
.26
Self-efficacy
.33
.35
.16
.23
.19
General mental ability
Conscien-tiousness
Emotional stability
Extraversion
Experience
13
Results Full Model with Distals
Self-set goals
.02
Job/Task Performance
.50
.02
Self-efficacy
.31
.16
.02
.31
.54
.33
.35
.16
.23
.19
General mental ability
Conscien-tiousness
Emotional stability
Extraversion
Experience
14
Role of Job Complexity
Notes Estimates are path coefficients. p lt
.05. p lt .01.
15
Other Moderating Influences
Notes p lt .01.
16
Results SUMMARY

• In the overall analysis, inclusion of distal
  variables undermines effects of self-efficacy on
  performance
• Relative impact of self-efficacy, though, depends
  on situation
• Positive in low complexity jobs/tasks
• Nil in medium complexity jobs/tasks
• Negative in high complexity jobs/tasks
• Measure also matters

17
Discussion

• Why the negative effect in high complexity jobs?
• Resource allocation theory (Kanfer Ackerman,
  1989) predicts that self-regulation steals
  valuable cognitive resources from complex skill
  acquisition
• However, multiple-resource models have been
  criticized (Neumann, 1987) and it does not appear
  that self-regulation requires significant
  attentional resources (DeShon, Brown, Greenis,
  1996)

18
Implications

• Is self-efficacy epiphenomenal?
• In some situations, yes
• In some situations, no
• Upshot
• Realize that in some situations self-efficacy may
  be unimportant or even detrimental
• Other moderators will be studied
• Feedback

19
Eds Comments (EEd, TTim)

• Stajkovic has a new MA of group efficacy
  studies--they come out as well as the individual
  studies
• We did not study group efficacy (but could if
  there were a sufficient number of correlations)
• Logically, distal variables should work thru
  proximate variables--did you try actual mediation
  or partial r tests?
• This may seem logical, but our results support a
  partially mediated model (indirect and direct
  effects of the distal variables) a fully
  mediated model was not supported by the results

20
Eds Comments (Continued)

• What does the model look like using only SE
  measured quantitatively (i.e., confidence summed
  over a series of performance outcomes after Ss
  had some task feedback)?
• These are the results for the grid measures
  (which indeed do suggest higher validities) if
  the point is that SE is only meaningful when a
  grid measure is used and with feedback, then one
  needs to confine the generalizations of SE to
  these situations

21
Eds Comments (Continued)

• Did you only include studies that included ALL of
  your variables?
• There are no studies that include all these
  variables, which is why path analysis of
  meta-analytic data is used, as it has in many
  recent studies
• There are thousands of SE studies but you have
  only a very small sample (and dozens of studies
  showing actual causal effects)
• We include roughly the same number of
  SE-performance studies as Stajkovic and Luthans
  studies of causal effects do not have the same
  variables as this study

22
Eds Comments (Continued)

• When you put goals and SE together in a model,
  they steal variance from each other because of
  being highly correlated
• This is certainly something we can look at
  (though shouldnt both be in causal model?)
• The r's for the big 5 seem much higher than most
  meta analysis have shown (usually measn r is
  about .20 isn't it?)
• The personality validities are from a
  meta-analysis of existing meta-analyses (Barrick,
  Mount, Judge, 2001) for conscientiousness, the
  validity is the same as Mount and Barrick (1995)

23
Eds Comments (Continued)

• SE cannot have negative effects on complex
  tasks--this makes no sense--and goal effects are
  smaller on complex tasks not stronger--maybe due
  to inclusion of goals with SE
• There may be a suppression effect (we can
  eliminate goals from model)
• In your search results table, you only have
  SE-GMA, SE-Exp, and goals only once--how did you
  fill your  correlation matrix with so much data
  missing???
• We relied on existing meta-analyses for the other
  bivariate relations (see slide 7)

24
Eds Comments (Continued)

• E The self set goal perf. mean r is unusually low
  isn't it (see the meta analyses in our
  book)--this means it may not be a representative
  sample of studies
• We relied on the most thorough meta-analysis on
  the validity of self-set goals (Harkins Lowe,
  2000) the validity of self-set goals is lower
  than for assigned goals
• We would be happy to use a different
  meta-analytic result if we felt it was more valid