If at First You Don't Succeed, Try, Try Again: Effects of Persistence

Journal of Applied Psychology 1988, Vol. 73, No. 2,208-216

Copyright 1 988 by the Am rican Psychological Association, Inc. 002l-90lO/88/$00.75

If at First You Don't Succeed, Try, Try Again: Effects of PersistencePerformance Contingencies, Ego Involvement, and Self-Esteem on Task Persistence Lloyd E. Sandelands and Joel Brockner Graduate School of Business, Columbia University

Mary Ann Glynn Yale University

We explored some of the factors affecting individuals' decisions to persist with a course of action. A total of 60 graduate students of varying levels of chronic self-esteem worked at a task that contained several insoluble problems (unbeknownst to the participants). One half were informed beforehand that the nature of the task was such that persistence was a wise strategy for task completion (continuous condition), whereas the remaining half were informed that the nature of the task was such that persistence was a less prudent strategy (discrete condition). Also, one half were told that their task performance was very revealing of their personality and aptitude levels (high-involvement condition), whereas the remaining half were informed that their task performance was nonrevealing of themselves (low-involvement condition). Subjects exhibited greater persistence in the continuous than discrete condition; the continuous-discrete difference was much greater in the high-involvement than low-involvement condition; and the continuous-discrete information had a greater impact on the degree of persistence exhibited by high than low self-esteem subjects.

Most organizational scholars and practitioners probably

and self-abusive at worst. In the popular idiom, persistent work-

would agree that commitment—and in particular its behavioral

ers are beating their heads against the wall in these instances.

concomitant of persistence at a course of action—often is desir-

Both their own and the organization's time would be better

able in work organizations. Indeed, there is ample empirical,

spent on activities that do promote organizational goals.

theoretical, and anecdotal evidence to support this assertion.

As the preceding comment implies, one of the most intrigu-

Staw and Ross (1980), for example, demonstrated that organi-

ing aspects of persistence is that individuals may go beyond rea-

zational decision makers were evaluated much more favorably

sonable bounds when committing themselves to courses of ac-

to the extent that they persisted with, rather than strayed from, a particular course of action. The vast psychological literature

tion. It has become fashionable among organizational scholars to explain such escalating commitment through mechanisms of

on learned helplessness (cf. Abramson, Seligman, & Teasdale,

self-justification (Festinger, 1957). That is, in order to convince

1978), which has been applied to the organizational context

themselves or important others that prior resources were not

(Martinko & Gardner, 1982), also implicitly values persistence.

allocated in vain, individuals, groups, and even entire organiza-

Learned-helplessness theorists argue that to give up in the face of negative feedback is individually and organizationally mal-

tions may persist with the failing course of action (Brockner & Rubin, 1985; Conlon & Wolf, 1980; Ross & Staw, 1986; Staw,

adaptive, especially if the negative feedback were due to external, unstable, and specific causes (Peterson & Seligman, 1984).

1976,1981; Teger, 1980). (For some recent alternative interpretations to the self-justification explanation of escalating com-

The virtue of persistence has even made its way into common

mitment, see Bowen, 1987, and Whyte, 1986.) Although self-justification notions may offer some insight

wisdom, perhaps best reflected in the age-old adage, "If at first

into the nature (and possibly dysfunctional consequences) of

you don't succeed, try, try again." Although persistence often is adaptive, it is not always adap-

persistence, we believe that these theories do not account for all

tive (i.e., if desired goals are unattainable). For example, suppose that workers lacked the ability for successful task perfor-

(or nearly all) of the variance in workers' persistence behavior. Rather, persistence may be equally—if not more so—a function

mance, or that the task was so difficult as to make its accom-

of the beliefs that people maintain about the nature of the rela-

plishment remote. In these cases, persistence is pathetic at best

tion between persistence and successful task performance. Expectancy theory (Vroom, 1964) posits that workers will be more likely to persist at a given task if they believe that successful performance is contingent, rather than noncontingent, on their

The first two authors contributed equally to this research. A version of this article was presented at the 1986 meeting of the Academy of Management in Chicago. We are grateful for the comments of anonymous reviewers on previous versions of the article. Correspondence concerning this article should be addressed to either Lloyd E. Sandelands or Joel Brockner, Graduate School of Business, Columbia University, New York, New York 10027.

level of persistence. Our study generally is grounded in this basic tenet of expectancy theory, although it is designed to explore three more specific questions: (a) Given that workers believe that there is some sort of positive relation (or contingency) between persistence and successful task performance, do various types of persistence-performance

contingencies differentially

affect persistence? (b) If so, does information about the nature

208

DETERMINANTS OF PERSISTENCE of the task influence workers' perceptions of the type of contingency between persistence and performance, and thereby in-

209

tasks evoke the perception that the contingency between persistence and performance is discrete.

fluence their persistence? Moreover, (c) what are some of the

In the present study, we manipulated participants' percep-

factors that moderate the impact of the perceived nature of the

tions of the contingency between persistence and successful per-

task on individuals' actual level of persistence?

formance by providing them with different information about the nature of the task. All of the participants were told beforehand that they were likely to find several trials of the task to be

Persistence-Performance Contingencies

quite difficult; in reality, although unbeknownst to participants, these problems were insoluble. One half of the participants

Although workers often believe that there exists some sort of

were informed that the task was one in which success was the

positive relation between task persistence and performance suc-

result of gradual, continuous effort (continuous condition),

cess, the precise nature of this positive relation can vary dra-

whereas the other half were told that task success was more the

matically. Thus, in certain circumstances, employees may be-

result of sudden insight than gradual persistence (discrete condition).

lieve that the persistence-performance relation is not only positive, but also relatively smooth and continuous. In this view,

In addition to testing the prediction that persistence would

task success is perceived to be the inevitable result of gradual, continuous effort devoted to the task. Moreover, workers have

be greater in the continuous than in the discrete condition, we wished to delineate several factors that might moderate the im-

the belief that they are drawing ever closer to their goals as they

pact of this task-information variable on persistence. The main-

allocate greater effort toward the task. We refer to these situa-

effect prediction—that persistence would be greater in the con-

tions as continuous, in that the positive relation between persis-

tinuous than in the discrete condition—may oversimplify the

tence and performance is perceived to be a continuous one.

relation between task information and persistence. In reality,

In other situations, workers may maintain generally positive

we suspect that the impact of the former on the latter is moder-

contingencies between persistence and performance; however, the specific form of that positive contingency may differ from

ated by a host of situational and dispositional variables. Moreover, expectancy theory and the self-esteem literature (discussed

the one previously described. That is, in many situations, suc-

ahead) may serve to guide the choice of moderator variables. In

cess results primarily from sudden insight into the task. In such

short, to maximize the theoretical or practical significance of

situations, individuals may devote considerable effort to the

the hypothesized relation between task information and persis-

task, which may or may not increase the likelihood of successful

tence, it is crucial to identify the conditions under which (or

performance. We refer to these situations as discrete, in that the

the individuals for whom) the relation is more or less true. Two

positive relation between persistence and performance is per-

moderator variables were explored in the present study: ego involvement and chronic self-esteem.

ceived to be a discrete one. From expectancy theory, it can generally be predicted that

Ego involvement. It is one thing for individuals to maintain

employees will exhibit greater task persistence when they be-

beliefs about the relation between persistence and successful

lieve that the persistence-performance contingency is continu-

performance; it may be quite another thing, however, for them

ous rather than discrete. In the case of the continuous contin-

to translate such beliefs into actual persistence behavior. Ac-

gency, workers are operating under the assumption that they are

cording to expectancy theory (e.g., Feather, 1982; Rubin &

moving inexorably closer to their goal as they persist longer at

Brockner, 1975; Vroom, 1964), individuals will be more likely

the task; much theory (e.g., Vroom, 1964) and empirical re-

to act on their beliefs to the extent that they have some incentive

search (e.g., Rubin & Brockner, 1975) have shown that individ-

to do so. In the present study, we manipulated participants' in-

uals are likely to persist with a goal-directed course of action to

centive to perform well, such that some had greater incentive

the extent that they are nearing the desired goal. In the case of the discrete contingency, workers are less convinced that they

(high-involvement condition) than others (low involvement). A

are moving closer to the goal as they persist longer. Therefore,

Task Information X Ego Involvement interaction effect was predicted: Persistence should be greater in the continuous than

we expected to find greater persistence when the positive contin-

in the discrete condition, particularly in the high-involvement

gency between persistence and performance was perceived to be

condition. Chronic self-esteem.

continuous rather than discrete. Many factors can influence workers' beliefs about the contin-

An additional purpose of the present

study was to explore a dispositional moderator of the impact of

uous versus discrete nature of the persistence-performance re-

the task-information factor on persistence. Theory and empiri-

lation, not the least of which is the perceived nature of the task.

cal research suggest that individuals' level of chronic self-es-

For some tasks, performance is primarily perceived to be effort

teem is one such factor. However, the available evidence from

dependent, such as in onerous but cognitively simple tasks that

the self-esteem literature makes conflicting predictions about

require stick-to-itiveness rather than any special intellectual

whether it is the individuals who are low in self-esteem (low SEs)

ability. Effort-dependent tasks evoke the perception that the

or the individuals who are high in self-esteem (high SEs) who

contingency between persistence and performance is continu-

are more apt to be affected by the task-information variable.

ous. On other tasks, performance is perceived to be much less

On the one hand, it could be argued that low SEs will be more

effort dependent. For example, insight tasks are those in which

influenced. Low SEs have little confidence in their own beliefs

individuals, on the one hand, may suddenly see the solution; on

in general (and about the perceived relation between persistence

the other hand, individuals also may stare at such tasks for long

and performance in this specific instance); therefore, they may

time periods without moving any closer to their goals. Such

be more influenced by external cues for appropriate behavior

210

L. SANDELANDS, J. BROCKNER, AND M. GLYNN

such as task information. In a study related to the present one, McFarlin, Baumeister, & Blascovich (1984) advised one half of the subjects to persist and the remaining half to move on if they did not solve the problem within a few moments. Low SEs were much more likely to follow this advice than were their high selfesteem counterparts. On the other hand, Janoff-Bulman and Brickman (1982) have provided theory and research to suggest that high SEs generally make better use (than do low SEs) of cues telling them whether it is prudent to persist or withdraw. This could be due to high SEs' greater confidence in being able to enact the behavior that such cues (as the task-information variable) deem appropriate. A more recent study by McFarlin (1985) lends empirical backing to this assertion. Subjects of varying levels of self-esteem worked on a variety of tasks, some of which—unbeknownst to them—were insoluble. Three experimental conditions were created. One group was told that all of the problems were soluble (soluble condition); a second group was told that some of the problems were insoluble (insoluble condition); and a third group was given no information at all. The primary dependent measure was subjects' persistence at the insoluble tasks. High SEs' level of persistence was much more affected by the solubility information than was low SEs' level of persistence. In particular, high SEs persisted far longer in the soluble than in the insoluble condition, whereas the low SEs did not. This finding is apparently inconsistent with the earlier McFarlin et al. (1984) result, in that it implies that high SEs will be more influenced by the task-information variable than will their low self-esteem counterparts. One purpose of the present study, therefore, was to evaluate the competing predictions concerning the nature of the Task Information X Self-Esteem interaction. Summary.

The present study consisted of a 2 X 2 X 2

between-groups design: Task Information (continuous vs. discrete) X Ego Involvement (high vs. low) X Self-Esteem (high vs. low), in which the primary dependent measure was task persistence. In addition to the aforementioned main effect for the task-information variable, we expected to obtain a significant Task Information X Ego Involvement interaction effect. Subjects should exhibit greater persistence in the continuous than in the discrete condition, and this tendency should be especially heightened if ego involvement is high rather than low. Because previous theory and research were equivocal about whether low SEs or high SEs would be more influenced by the task-informa-

problem solving by use of a computer, rather than by a more traditional pencil-and-paper format. Subjects were told that because computers have become quite common in the workplace, it is important to understand their impact on how individuals solve problems. Upon arrival at the business school's computer laboratory, subjects were greeted by the experimenter and seated before an IBM PC microcomputer that they used for the task. Prior to beginning the task, all of the subjects completed the Revised Janis-Field Self-Esteem Scale (Eagly, 1967). In a review of the psychometric qualities of existing self-esteem scales, Robinson and Shaver (1973) reported that this was one of the best scales to use with normal adults. Perhaps the best evidence of the measure's construct validity stems from research conducted subsequent to the Robinson and Shaver review. Specifically, Bruckner and his colleagues have used this scale in 15-20 studies exploring the relation between self-esteem and a wide variety of dependent measures (e.g., task performance, cognition, and affect). In each of the experiments, low SEs (defined as those scoring in the bottom portion of the overall distribution) differed from the higher SEs (denned as those scoring in the top portion of the overall distribution) on the dependent variable(s) of interest in the theoretically expected direction (see Brockner, Davy, & Carter, 1985; Brockner & Elkind, 1985; and Brockner & Swap, 1983, for a representative sampling of these 15-20 studies). A median split was used in the present study to classify subjects as high or low in self-esteem. The task consisted of 20 six-letter anagrams. These were presented on the computei; one at a time. Subjects had to enter their responses by typing in letters on the keyboard in front of them. Once they believed that they had successfully solved the anagram, they were to enter it and then move on to the next one in the sequence. They could also move on to the next one if they could not solve the anagram. Once having moved on, however, they were not able to return to an earlier problem in the sequence. Unbeknownst to subjects, four of the anagrams were insoluble. For each participant, we computed the mean time spent on the insoluble problems; this served as the primary measure of persistence. All of the participants were provided 20 min to work on the problems. To control for possible order effects, two different sequences were used. (One sequence was the reverse order of the other.) Subjects of varying levels of self-esteem were then randomly assigned to one of four conditions created by the 2 X 2 factorial crossing of the ego-involvement and taskinformation variables. The manipulated variables were operationally denned on the computer screens facing subjects. Ego involvement. Subjects' ego involvement in task performance was manipulated by the description of the task that all of the subjects had read prior to working on the task. Subjects in the high-involvement condition had read the following:

tion variable, the nature of the Task Information X Self-Esteem interaction was left unspecified at the outset.

Method Participants A total of 60 master of business administration students at Columbia University voluntarily took part in this laboratory experiment. Their mean age was 26 years, and they reported having an average of 4 years of work experience.

Procedure Subjects were run one, two, or three at a time. Many of the procedures of the study were automated through the use of computers. Indeed, the study was presented to participants as one that investigated

This task has been given to many people from all walks of life, and has been used to make numerous important judgments about people. Although it seems reasonably straightforward, it actually measures a number of significant traits. For example, it is a good indicator of a person's verbal intelligence. In addition, it is very revealing of an individual's ability to see the relationships between letters and words, which in turn is highly related to a variety of perceptual and intellectual skills. Also, it has been shown that a person's ability to solve this task with the use of a computer (rather than paper and pencil) measures problem solvingflexibility,a trait that is obviously very important to have. Finally, it should be mentioned that many employment agencies and businesses use tests very similar to this one when making hiring decisions. It is very typically found that those who perform well on this type of task tend to be the most satisfied and productive employees. In short, high scores on this test are positively correlated with verbal intelligence, perceptual and intellectual abilities, and job productivity and satisfaction.

DETERMINANTS OF PERSISTENCE Subjects in the low-involvement condition read the following description of the task: This task has only recently been developed, and tested on just a handful of people. Consequently, its usefulness in making judgments about people is very much in doubt at this point. For example, it is not related to verbal intelligence, perceptual abilities, or anything of that sort. To our knowledge, moreover, the task has not been used in business or industry. In short, high scores on the task are not necessarily reflective of anything about the individuals who are attempting to solve it, and this is true regardless of whether the task is given on computer, as we are doing today, or with the more traditional paper-and-pencil procedure. Task information. This manipulation took place shortly after the ego-involvement manipulation. All of the subjects were told that the puzzles varied in difficulty. One half of the subjects were informed that the nature of the task was such that success depended on gradual, continuous persistence (continuous condition). They read the following ad-

Experience shows that if you are having difficulty with some of the harder ones, then you should stick with them and try to solve them. Generally speaking, this is the kind of task in which the longer you stick with it, the better chance you have of solving the task. Of course, if you still cannot solve a particular puzzle, then go on to the next one. But, it is probably best to do that when all else fails. The remaining one half of the subjects were given different information about the nature of the task (discrete condition). They read the following: Experience shows that if you are having difficulty with some of the harder ones, then you should move on fairly quickly to the next one. Generally speaking, this is the kind of task in which if you don't see the solution fairly quickly, you might not ever see it at all. Of course, you should still devote some time to solving each puzzle, even if you are having difficulty. But, in general, it is probably best to move on quickly if all else fails. Dependent variables. Behavioral measures of task performance were obtained by programming the computer to record both the amount of time subjects spent on each puzzle and the correctness of the solution that was entered for each puzzle. Persistence was measured as the mean amount of time spent on the insoluble anagrams. We also assessed the amount of time that subjects devoted to the soluble anagrams (both those that they correctly solved and those that they did not). However, the analyses focused on the insoluble anagrams because these items furnished the purest measure of persistence. Note that for the soluble anagrams, persistence could be measured only up to the time that the anagram was solved, after which the participant could not persist any further. This feature did not exist, of course, in the case of the insoluble anagrams. To calculate persistence, we divided the total time spent on the insoluble anagrams by the actual number of insoluble puzzles that the subject encountered. The insoluble puzzles were Numbers 5,9,13, and 16 in one order, and Numbers 5,8, 12, and 16 in the other order in the sequence of 20 puzzles. Thus, the maximum number of insoluble puzzles that any subject encountered was four. (We chose not to bunch up the insoluble anagrams in the first half of the task, in order to avoid giving subjects the impression that all [or most] of the anagrams were insoluble. That is, if more insoluble trials appeared early in the sequence, subjects may have concluded erroneously that all of the trials were insoluble, thereby greatly invalidating the behavioral measure of persistence.) We also assessed subjects' performance on the soluble problems (e.g., absolute number and percentage of correct solutions). After the anagrams task, all of the participants completed a question-

211

Table 1 Persistence" as a Function of Ego Involvement and Task

Information Task information

Ego involvement

Low M SD n High M SD n

Continuous

Discrete

171.58 56.01 15

165.35 85.93 15

186.39 90.89 15

114.16 48.34 15

" Measured as the mean time (in s) spent on insoluble puzzles. Higher values indicate greater persistence.

naire that included an ego-involvement manipulation check. Specifically, subjects rated the extent to which the task was "involving" on a 4-point scale ranging from not at all ( I ) to very much (4). They also evaluated the task, along 4-point scales, on nine other dimensions from the Work scale of the Job Descriptive Index (JDI; Smith, Kendall, & Hulin, 1969): fascinating, routine, satisfying, boring, good, creative, pleasant, challenging, and simple. Endpoints were not at all (1) and very much (4). Subjects were asked the following questions: (a) "All in all, how would you rate this exercise?" Endpoints were a lot of fun (9) and not at all fun (I). (b)" All in all, how would you rate your performance of this task?" Endpoints were very good (9) and very poor (1). They were also asked how much they agreed with the statement, "I put a lot of effort into performing well in this exercise." Endpoints were strongly agree (1) and strongly disagree (5). Subjects were asked to write down any suspicions they had about the study, as well as their hypotheses of its true purpose. Not one person was unduly suspicious, and no one indicated anything resembling the major hypotheses. Finally, all of the participants were fully debriefed and thanked for their participation. Results Unless otherwise specified, all of the data were subjected to a three-factor, unweighted-means analysis of variance (ANOVA). (Preliminary analyses revealed that the order variable produced no effects on the dependent variables; therefore, it will not be discussed further.)

Behavioral Data Number of insoluble anagrams attempted. We first sought to determine if there were condition differences in the number of insoluble anagrams attempted. The ANOVA revealed only a significant main effect for ego involvement, F(\,52) = 6.23, p < .02, such that participants attempted more insoluble problems in the high-involvement (M = 3.13) than in the low-involvement condition (M = 2.57). This finding is not of major theoretical importance, especially in light of the fact that our chief dependent measure—time spent on the insoluble problems—was determined by computing each participant's mean time spent on the insoluble problems. Persistence at insoluble problems. The ANOVA revealed a significant task-information main effect, F( 1, 52) = 4.79, p < .03,

212

L. SANDELANDS, J. BROCKNER, AND M. GLYNN

Table 2

the high-involvement condition (M = 44.58 s). For the latter,

Persistence" as a Function of Self-Esteem

there were significant ego-involvement and task-information

and Task information

main effects (both p values < .02). Subjects spent more time on the soluble puzzles that they could not solve in the low-involve-

Task information Self-esteem High M SD n Low M SD n

ment than in the high-involvement condition (Ms = 151.94 s

Continuous

Discrete

199.79 76.88 15

111.53 37.54 13

158.18 68.38 15

161.32 86.92 17

vs. 105.26 s) and in the continuous than in the discrete condition (Ms = 150.82 s vs. 106.38 s). Thus far, we have reported two findings that suggest that the task-information variable had its intended effect: (a) the main effect for this factor on the major dependent variable (i.e., persistence at the insoluble anagrams) and (b) the main effect for

" Measured as the mean time (in s) spent on insoluble puzzles. Higher values indicate greater persistence.

this variable on the amount of time subjects devoted to the soluble anagrams that they did not solve. In both instances, subjects persisted significantly longer in the continuous than in the discrete condition. Further evidence on the efficacy of the taskinformation variable stems from an analysis of the total number of anagrams attempted. If subjects were responsive to the task-

a 2 = 05. As expected, subjects persisted for a longer period of

information variable, then they should have attempted to solve fewer anagrams in the continuous than in the discrete condi-

time in the continuous (M = 178.99 s) than in the discrete con-

tion, by virtue of the fact that they spent more time on certain

dition (M = 139.75 s). This effect was complemented by two

anagrams in the former than the latter condition. Of course, the

significant interaction effects. First, as predicted and can be

number of anagrams correctly solved correlated positively with

seen in Table 1, subjects' degree of persistence was much more sharply influenced by the task-information variable in the high-

the total number attempted, r(58) = .82, p< .001. Therefore,

involvement than in the low-involvement condition, F( 1,52) =

ber of anagrams attempted; the number of anagrams correctly

4.52, p < .04, a2 = .05. Simple effect analyses revealed that

solved served as the covariate. As predicted and can be seen in

we performed an analysis of covariance (ANCOVA) on the num-

within the high-involvement condition, subjects persisted sig-

Table 4, subjects attempted significantly fewer anagrams in the

nificantly longer in the continuous than in the discrete condi-

continuous than in the discrete condition (adjustedMs = 12.79 vs. 14.23, respectively), F( 1,51) = 6.43, p < .02. Taken together,

tion, F(\, 52) = 7.40, p < .01, w 2 = .09; no such difference emerged in the low-involvement condition, F