Revisiting the Functional Properties of Self-Efficacy

Social Cognitive Theory and Levels of Analysis

Social cognitive theory has been, and will continue to be, influential in the field of management and related disciplines. It is widely accepted that self-efficacy plays a critical role in self-regulation (e.g., Schmidt, Beck, & Gillespie, 2013; Sitzmann & Ely, 2011). There is overwhelming evidence, from both correlational and experimental studies, showing that highly efficacious individuals tend to set more challenging goals (e.g., Robbins, Lauver, Lee, Davis, Langley, & Carlstrom, 2004; Sitzmann & Ely, 2011), strive harder (e.g., Colquitt, LePine, & Noe, 2000; Sitzmann & Ely, 2011), persist for longer (e.g., Multon, Brown, & Lent, 1991), feel better (e.g., Judge & Bono, 2001; Sitzmann & Ely, 2011), and achieve better results than their counterparts (Moritz, Fleltz, Fahrbach, & Mack, 2000; Multon et al., 1991; Sadri & Robertson, 1993; Stajkovic & Luthans, 1998). However, these meta-analyses, as well as the majority of primary studies included in them, were conducted at the between-person level of analysis, which does not provide a complete picture of self-efficacy’s role in self-regulation. Lord, Diefendorff, Schmidt, and Hall (2010), for example, observe that although “between-person research consistently shows positive associations between self-efficacy and performance . . . within-person research finds self-efficacy can be negatively associated with subsequent performance” (pp. 544-545) and argue that between-person research “may do little to inform our understanding of within-person self-regulatory processes” (p. 545).

The problem that Lord et al. (2010) referred to is known as the “ecological fallacy” (Chen, Bliese, & Mathieu, 2005; Cooper & Patall, 2009; Robinson, 1950)—it involves equating findings that are based on aggregated data to findings based on disaggregated data that ignore group membership. Robinson’s (1950) seminal work demonstrated that literacy rates were higher in U.S. states with a greater percentage of foreign-born individuals (r = .62) but that literacy was negatively correlated with foreign birth when the data were examined at the individual level (r = –.12). This example of the ecological fallacy involves individuals nested within groups; however, the ecological fallacy also applies to situations in which repeated measurements are nested within individuals (i.e., the individual is the “group”). For example, individuals who exercise frequently tend to have lower blood pressure than those who exercise less frequently; yet for any given individual, an increase in exercise can increase blood pressure. ¹ As these examples show, it is essential to design and analyze effects at the level at which they are expected to operate.

Social cognitive theory describes a set of processes that operate at the within-person level of analysis. Self-efficacy is defined as an individual’s belief regarding his or her capability to succeed and attain a given level of performance (Bandura, 1977). Social cognitive theory identifies two mechanisms by which this belief influences motivation. Bandura (1977, 1997) terms these mechanisms “discrepancy production” and “discrepancy reduction.” The argument is that high self-efficacy causes people to set more challenging goals, thereby creating a discrepancy between the current and desired state. The person then acts to reduce this discrepancy by applying more effort and using more effective strategies. These are dynamic processes that unfold within individuals, over time, and should thus be observed and analyzed at the within-person level.

Although Bandura (2012) identified a number of requirements for the test of social cognitive theory, he omitted one important condition, namely, that any empirical test of a theory must be conducted at the appropriate level of analysis. A problem with empirical tests of social cognitive theory is that, until recently (e.g., Vancouver, Thompson, & Williams, 2001), most studies analyzed effects at the between-person level rather than the within-person level. The issue of levels of analysis is central to the debate over the functional properties of self-efficacy, because mis-specification of levels may produce spurious results (Sitzmann & Yeo, in press). Aggregating within-person variance to the between-person level can mask the true direction and/or magnitude of effects at the within-person level. Aggregation is particularly problematic for situations in which a given construct (such as self-efficacy) could have opposing effects across levels of analysis. Analyses conducted at the between-person level are useful for addressing research questions that relate to static differences between individuals. For example, such analyses were appropriate for Bandura’s early work that examined whether individuals with higher self-efficacy were more likely to overcome phobias than those with lower self-efficacy (e.g., Bandura & Adams, 1977; Bandura, Reese, & Adams, 1982). However, between-person analyses in isolation cannot assess how changes in self-efficacy within an individual relate to subsequent changes in performance. In the next section, we demonstrate how the concept of resource allocation can explain the co-existence of positive and negative dynamic self-efficacy effects across the between- and within-person levels of analysis.

Resource Allocation Theory and the Direction of Dynamic Self-Efficacy Effects

Bandura accused us of piecemeal theorizing. Needless to say, we disagree. We used a single mechanism—cognitive resources—to generate predictions regarding the way in which the relationships among self-efficacy (both general and task specific) and performance may change throughout skill acquisition. There is a long history in psychology of using the hypothetical concept of cognitive resources (Humphreys & Revelle, 1984; Kahneman, 1973) as an explanatory mechanism. The concept of cognitive resources is useful, because it provides a means of explaining complex phenomena that would otherwise be difficult to account for. Examples include the effects of dual-task interference, arousal, fatigue, ability, and motivation on performance (Hockey, 1997; Humphreys & Revelle, 1984; Kanfer & Ackerman, 1989; Navon & Gopher, 1979; Yeo & Neal, 2004).

In the current context, the concept of cognitive resources provides a means of understanding the effects of self-efficacy on performance in a dynamic context—that is, in which both constructs change over time and thus vary within, as well as between, individuals. Changes in resource demands over time can be used to explain how the effects of self-efficacy at different levels of analysis may change throughout skill acquisition. Previous results regarding the relationship between self-efficacy and performance over time have been mixed—the relationship strengthened in some studies yet weakened or remained stable in other studies (Lee & Klein, 2002; Mitchell, Hopper, Daniels, George-Falvy, & James, 1994; Vancouver et al., 2001). Resource allocation theory (Kahneman, 1973; Kanfer & Ackerman, 1989; Yeo & Neal, 2004) provides a principled basis for reasoning about these effects. The core elements of this approach are explained below.

Following Kanfer and Ackerman (1989), we assume that individuals have a limited pool (or availability) of cognitive resources, which can be allocated to on-task, off-task, and self-regulatory activities. Resource availability is expected to differ across individuals, primarily as a function of cognitive ability. The greater the pool of cognitive resources, the more capacity an individual has for setting higher goals, persisting for longer, and using more effective strategies than his or her counterparts. Resource allocation is expected to vary over time, due to dynamic contextual factors, such as task demands and task engagement. The allocation of these limited resources is governed by a compensatory mechanism such that resources are allocated according to need (Hockey, 1997; Vancouver, 2005). For example, when task demands are low, an individual will devote relatively few resources to the task. As task demands increase, the individual will increase his or her allocation of resources to the task, until he or she reaches the point of maximum motivational potential, which is the maximum amount of effort that the individual is prepared to allocate to the task (Brehm & Self, 1989; Hockey, 1997). At this point, the person may change his or her strategy, lower his or her goal, or withdraw from the task.

Turning to the effects of self-efficacy, prior research suggests that people who are highly efficacious tend to have higher ability, use more effective strategies, set more difficult goals, and persist for longer than people who are not efficacious (Chen, Gully, & Eden, 2004; Kanfer & Heggestad, 1997). ² If so, then people with high self-efficacy should, on average, allocate more resources to the focal task, and use those resources more effectively, than people with low self-efficacy. As a result, people with high self-efficacy should learn faster, and reach their performance asymptote earlier, than people with low self-efficacy. These processes should produce a positive main effect of self-efficacy on performance at the between-person level that initially strengthens over time as the performance trajectories diverge and then weakens as the performance trajectories converge (after individuals with high self-efficacy reach their asymptote).

At the within-person level, this account suggests that fluctuations in self-efficacy during a performance episode should be negatively associated with the dynamic allocation of resources. Efficacy judgments are a source of information indicating what proportion of resources are needed to meet task requirements—a decrease in self-efficacy signals the need for more resources to compensate for higher task demands (Vancouver, Weinhardt, & Schmidt, 2010). Consequently, self-efficacy should be negatively associated with performance at the within-person level. Note that although self-efficacy and performance trajectories are expected to be positive overall (as described above), dynamic contextual factors (e.g., task demands, task engagement) should produce within-person deviations around these trajectories (see also Sitzmann & Yeo, in press, for a description of this phenomenon). For example, when task demands reduce from one trial to the next, an individual’s self-efficacy should increase. This change in self-efficacy should prompt a reduced allocation of cognitive resources and subsequent decrease in performance. When task demands increase from one trial to the next, an individual’s self-efficacy should decrease. This change in self-efficacy should prompt an increased allocation of cognitive resources and subsequent improvement in performance.

Our results (Yeo & Neal, 2006) are consistent with this theoretical account. Task-specific self-efficacy was negatively related to performance at the within-person level, whereas general self-efficacy and average task-specific self-efficacy were positively related to performance at the between-person level. Furthermore, the average level of task-specific self-efficacy reported by participants during the experiment predicted the shape of the performance trajectory. This interaction is particularly noteworthy—as shown in Figure 1, the performance trajectories of individuals with high self-efficacy diverged from those with low self-efficacy early in practice but then converged again toward the end as individuals with high self-efficacy reached their performance asymptote. This finding is consistent with our theoretical framework and shows why it is important to consider the dynamics of the underlying process when reasoning about the effects of self-efficacy.

OPEN IN VIEWER

Figure 1

Cross-Level Interaction Between Average Task-Specific Self-Efficacy and Practice

Bandura (2012) questioned how it is possible for general self-efficacy and task-specific self-efficacy to work in opposing directions. We are puzzled by this reaction because the notion of opposing self-efficacy effects is consistent with social cognitive theory. As explained above, the role of discrepancy production processes in generating positive self-efficacy effects is central to social cognitive theory (Bandura, 1977). In parallel, social cognitive theory also recognizes the role of discrepancy reduction processes and acknowledges that self-efficacy effects may be negative when such processes are active (e.g., Bandura & Locke, 2003).

It is important to note that there are boundary conditions relating to the effects described above. The theoretical explanation that we have presented assumes that people act to conserve resources. This mechanism is important in situations in which people have competing demands and limited resources, but it is not universal. For example, many of the early examinations of self-efficacy used tasks or settings in which people were asked to cope with something aversive (e.g., submersing one’s arm in freezing cold water; Litt, 1988) or interact with feared stimuli (e.g., snakes for people with a snake phobia; Bandura & Adams, 1977). These tasks do not require people to make choices regarding the allocation of a limited pool of resources. Thus, an increase in self-efficacy in these tasks is unlikely to signal that the person can reduce the amount of effort that they are applying.

Sitzmann and Yeo’s (in press) within-person meta-analysis highlighted the importance of boundary conditions. Their results showed a null relationship between self-efficacy and performance after controlling for past performance and the linear trend. However, substantial variability was unaccounted for by available moderators, so they urged researchers to continue searching for moderators to understand when the effect may be positive or negative. Vancouver, Schmidt, and their colleagues (Beck & Schmidt, 2012; Schmidt & DeShon, 2009, 2010; Vancouver, More, & Yoder, 2008) are leading the field in this regard. To date, the potential moderators assessed in these papers have been treated somewhat independently. However, an implicit theme underlying this work is that a negative self-efficacy effect is expected when an increase in self-efficacy signals that resources are not required for success and can thus be conserved (e.g., for people with high average self-efficacy, in easy tasks, and following prior success or satisfactory goal progress); whereas a positive effect is expected when an increase in self-efficacy signals that resources are likely to improve the chance of success and thus be a worthwhile investment (e.g., for people with low average self-efficacy, in difficult tasks, and following prior failure or unsatisfactory goal progress).

For example, Beck and Schmidt (2012) provided support for their argument that a negative effect exists for people with high average self-efficacy (those whose judgments of self-efficacy fluctuate around the high end of the continuum). The authors argued that these individuals believe that the task can be completed easily, so a momentary increase in self-efficacy for them indicates that resources can be diverted elsewhere. On the other hand, Beck and Schmidt provided support for their argument that a positive effect exists for people with low average self-efficacy. The authors argued that these individuals believe that success is in doubt, such that a momentary increase in self-efficacy for them should indicate that there may be a chance of success if resources are allocated to the task. Interpreting these findings under our framework suggests that both individuals with high and low average self-efficacy act to conserve resources but under different conditions and in different ways. Those with high average self-efficacy act under the premise that resources will not necessarily be required for success, so they will be expended only when required; otherwise they will be conserved for another task. Those with low average self-efficacy act under the premise that resources will be required for success, so they will use them when they are likely to lead to success and otherwise conserve them for another time during engagement of the current task.

Methodological Considerations

In addition to the criticisms addressed above relating to theory, Bandura (2012) also criticized our (Yeo & Neal, 2006) work on a number of methodological grounds. We address these comments below.

Conclusion

In conclusion, we are puzzled by Bandura’s reaction to our results, given that social cognitive theory allows for the possibility of negative effects of self-efficacy at the within-person level. It is hard to escape the conclusion that Bandura is being selective in his critique, seeking to criticize studies that demonstrate a negative within-person self-efficacy effect, without applying the same standards to within-person studies that demonstrate a positive effect of self-efficacy. For example, Bandura (2012) criticized our 2006 paper yet cited one of our unpublished conference papers (Neal & Yeo, 2003) as one of a number of “well-designed studies of the role of self-efficacy in within person change [that] verify its positive function” (p. 18). This paper was cited because it demonstrated that the relationship between self-efficacy and performance became positive when participants were provided with unambiguous feedback. We did attempt to publish this work; however, the paper was rejected for publication because the manipulation of task ambiguity was confounded. It seems ironic that work that has not passed the test of peer review is viewed more favorably than work that has.

It is important to note that we are not claiming that the negative effect of self-efficacy at the within-person level is universal (it is not), and we are not advocating social policies that undermine self-efficacy. Like others (e.g., Beck & Schmidt, 2012; Schmidt & DeShon, 2009; Seo & Ilies, 2009; Vancouver & Kendall, 2006), we are trying to identify boundary conditions that influence the direction and magnitude of self-efficacy effects on motivation and performance and to understand the mechanisms that underlie these dynamic processes. Over the last decade, researchers have made great strides in understanding this complex and dynamic process (Schmidt et al., 2013; Sitzmann & Yeo, in press). We have no doubt that our collective understanding of this process will continue to improve over the next decade and beyond, provided we continue to apply the resources that are required to achieve this goal.