Online Investment Self-Efficacy: Development and Initial Test of an
Instrument to Assess Perceived Online Investing Abilities
Clayton A. Looney
University of Virginia
[email protected]
Joseph S. Valacich
Washington State University
[email protected]
Asli Y. Akbulut
Grand Valley State University
[email protected]
Abstract
This paper develops and tests an instrument to measure online investment self-efficacy, defined as an individual's perceived ability to utilize online technologies to accomplish investing-related tasks. A series of empirical studies were conducted to establish the measure's psychometric properties. The results suggest that the measure exhibits admirable levels of reliability, as well as convergent, discriminant, and nomological validity. As predicted by theory, computer self-efficacy was found to serve as an important precursor to online investment self-efficacy. Furthermore, online investment self-efficacy played a significant role in fueling investor preference for the traditional (full-service) or online investing approach. More efficacious investors tended to prefer Web-based technologies as a vehicle for investing, whereas less efficacious individuals favored the traditional method.
1. Introduction
Undeniably, the Internet and World Wide Web (Web) have altered our society. We now communicate interactively and instantaneously over vast distances, receive a wide array of information tailored to our unique needs, and conduct our business remotely without human assistance. One such area that has been significantly impacted by this development is the investment of our financial resources. While in the past
individuals have relied on trained investment professionals to assist them in planning for their futures, today individual investors can take personal control over their financial destinies by investing their money using online investment firms.
Prior to the emergence of Web-based investing (circa 1993), the vast majority of individual investors worked with full-service brokerage firms such as Merrill Lynch, Morgan Stanley, or Prudential Securities. The firms' seasoned investment professionals offered personalized advice to individual investors, taking care of their financial planning needs by building and managing investment portfolios. This type of financial intermediary is referred to as
traditional investing. Unlike traditional brokerage
firms who largely control the investment activities of their clients, online brokerage firms such as E*Trade and Ameritrade have deployed Web-based technologies that disintermediate the trained financial expert, allowing individual investors to participate in financial exchanges directly without human assistance. This approach is referred to herein as online investing.
Even though online investing offers novel ways for individual investors to interact with financial markets, not all investors elect to utilize the Web as a vehicle for investing. Both full-service and online brokerage firms survive and thrive in today's financial industry [13]. Given this evidence, it appears that individual differences may play a key role in an investor's decision to invest online. However, little is understood about the underlying mechanisms that fuel investor preference for the traditional versus online
approach. A highly plausible explanation involves an individual's self-efficacy, which refers to a belief in one's ability to perform a particular behavior [2]. Self-efficacy has proven influential in performance attainments, effort expenditures, emotional reactions, and, particularly pertinent to the present effort, decisions regarding particular behaviors in which to engage [3].
As a first step in a systematic program of research, this paper focuses on the concept of online investment self-efficacy, which refers to an individual's perceived ability to utilize online technologies to accomplish investing-related tasks. To our knowledge, no existing instrument has been put forth to measure online investment self-efficacy. Hence, this research serves two important purposes. First, it develops a valid and reliable measure of online investment self-efficacy via a rigorous instrument development process. Second, it empirically examines the measure's ability to predict an important outcome related to the online investing phenomena, namely an investor's decision whether to engage in traditional or online investing.
To accomplish these objectives, the paper is organized as follows. The next section covers the theoretical underpinnings that serve as the cornerstone for this research. Based on a sound theoretical foundation, a research model is then offered, as well as a set of related hypotheses to assess its veracity. Then, the instrument development procedure is outlined and the results obtained from the analyses are presented. The paper concludes with a discussion of the implications flowing from this research and offers suggestions concerning the utility of the online investment self-efficacy measure in future research.
2. Theoretical Underpinnings
Social Cognitive Theory (SCT) [2] provides a useful meta-framework for investigating phenomena related to online investing. SCT has proven to be a powerful mechanism for explaining, predicting, and controlling behavior, receiving empirical support across a variety of domains including education, health, clinical psychology, athletics, and organizational functioning [3]. Specific to IS research, SCT has been successfully applied in studies related to training [1], technology acceptance [18], technology use [7], and virtual organizations [16] to name a few.
Referring to Figure 1, SCT views psychosocial phenomena as reciprocally influenced by environmental, personal, and behavioral factors. SCT is based on the premise of triadic reciprocality [2], which proposes that personal, environmental, and
behavioral factors are mutually and reciprocally determined. People enter contextual situations with a set of abilities, traits, histories, and cognitive resources to deploy during their interactions with the environment. Environmental forces such as regulations and innovations can restrict or enable certain types of behavior. When specific behaviors are enabled, individuals assess their ability to engage in these behaviors by integrating perceptions of themselves, the environment, and the particular behavior in question. Behavior in a given situation is, therefore, mutually determined by environmental and personal components. The environment can be transformed as a result of one's action and this behavior provides feedback to the individual, resulting in a reassessment of his or her capabilities as well as the nature of the environment.
Environment
Person Behavior
Figure 1. Social Cognitive Theory [2]
Based on individual factors, specific environmental conditions, and particular behavior being examined, the interactions between these factors will be different depending on the exact situation. According to the model, environmental, personal, and behavior factors can exert influence simultaneously and bi-directionally. However, the triadic system of reciprocality does not imply that the linkages exert equal influence or operate concurrently. In contrast, their relative importance will fluctuate depending on specific activities and situational circumstances [3]. Moreover, causal factors exert their influence over time. Beliefs crystallize based on repeated interactions within specific contexts.
Central to the conceptualization of the person component of SCT is the concept of self-efficacy, which refers to personal judgments of how well one can execute courses of action to accomplish a given task [2]. Collectively, IS research suggests that self-efficacy plays a critical role when one interacts with information technologies [1,7,11,14,16,18]. In the computing domain, computer self-efficacy can be defined as "an individual judgment of one's capability to use a computer" [7, p. 192], Computer self-efficacy has been found to exercise influence on an individual's
expectations, emotional reactions, and actual use of information technologies [7].
Self-efficacy judgments are found to vary along three distinct dimensions: generalizability, magnitude, and strength. Generalizability refers to the level of specificity for which the self-efficacy beliefs generalize across tasks and situations. Magnitude involves the particular level of task difficulty, whereas strength captures an individual's conviction of successfully performing at a particular level of task difficulty [3]. Assessing an individual's magnitude and strength are mainly measurement concerns, which are addressed in Methodology section. However, self-efficacy generalizability plays a key role in establishing self-efficacy judgments in the realm of online investing, as these tasks are theoretically related to other forms of self-efficacy.
Referring to Figure 2, self-efficacy generalizability can be viewed in a hierarchical manner, ranging in level of abstraction from general to domain to task. General self-efficacy, defined as "one's belief in one's overall competence to effect requisite performances across a wide variety of achievement situations" [5, p. 63], resides atop the hierarchy. This construct intends to capture an individual's perceived capabilities in a broad range of ability domains.
General
Task
Supported Not Yet Investigated
Online Investment Self-Efficacy Excel Self-Efficacy Windows 95 Self-Efficacy Domain G e ne ra liz a b ility L e ve l Computer Self-Efficacy General Self-Efficacy Leadership Self-Efficacy
Figure 2. Self-efficacy generalizability
Domain level self-efficacies are more specific, focusing on a broad range of tasks within a particular realm. For instance, in the computing domain, computer self-efficacy has been defined as "an individual's judgment of efficacy across multiple computer applications" [14, p. 129], whereas task level self-efficacy has been defined as "an individual's perception of efficacy in performing specific computer-related tasks within the domain of general computing" [14, p. 128]. Links have been established between general self-efficacy and ten domain level self-efficacies, including leadership self-efficacy [5]. Supporting the domain-task linkage in IS, studies have found computer self-efficacy to be a significant
predictor of task level forms, such as Excel self-efficacy [11] and Windows 95 self-self-efficacy [1].
Furthermore, efficacy theory posits that self-efficacy judgments should match the behaviors they intend to predict [2,3]. For instance, when examining an individual's use of Excel, Excel self-efficacy should be more predictive compared to computer self-efficacy since it is more intimately related to Excel behaviors. Furthermore, general self-efficacy should be the least predictive due to its theoretical detachment from the behavior in question.
According to the self-efficacy generalizability framework, online investment self-efficacy must be operationalized at the task level to enable prediction of online investing phenomena. Since online investment self-efficacy involves computing skills to achieve successful outcomes, the construct should be related to its domain level parent, computer self-efficacy. The link between general and domain self-efficacies, however, has yet to be established in the IS literature. Therefore, it is presently unclear how intimately related general self-efficacy beliefs are to those in the computing domain. Given the gaps in the literature, this research not only contributes to our knowledge through the development of a valid and reliable measure that is predictive of an important online investing phenomenon, but also fills the remaining voids in the generalizability framework. The next section details a comprehensive research model and a set of related hypotheses that enable these contributions to be brought to fruition.
3. Research Model and Hypotheses
The proposed research model (Figure 3) represents an important initial effort to assess the socio-technical phenomena of online investing from the SCT perspective. H2b Computer Self-Efficacy Online Investment Self-Efficacy H1 General Self-Efficacy Preference H3c H2a H3a H3b General Task Domain G e neraliz ab ilit y Lev el
Figure 3. Research model
As previously mentioned, general self-efficacy has been empirically linked to various domain level self-efficacies in organizational settings [5]. Given that
computer self-efficacy captures beliefs in the realm of computing at the domain level, the generalizability framework leads to the belief that the general-domain relationship should similarly hold in the realm of computing. Thus, the following hypothesis is offered.
H1: General self-efficacy will have a
significant positive influence on computer self-efficacy.
Domain level computer self-efficacy beliefs have been captured by operationalizing the construct as one's ability use a computer in the context of general application use [7]. Excel [11] and Windows 95 [1] self-efficacies have also been operationalized, capturing self-efficacy beliefs at the task level. Both measures provided empirical evidence for the domain-task level link within the computing domain by establishing connections between computer self-efficacy and the task level forms. Since the task of online investing requires abilities in the domain of computing, it logically follows that computer self-efficacy should predict online investment self-self-efficacy as well. However, the general to task level connection has not been empirically investigated in the literature, making it unclear whether its effect on online investment self-efficacy will be significant. Given that general self-efficacy is theoretically more distant than computer self-efficacy, its effect on online investment self-efficacy should be less pronounced. Hence, the following hypotheses are offered:
H2a: General self-efficacy will not have a
significant positive influence on online investment self-efficacy.
H2b: Computer self-efficacy will have a
significant positive influence on online investment self-efficacy.
Online investment self-efficacy should predict important outcome variables related to online investing, whereas general and domain level self-efficacies should be less predictive given their non-specific nature. Individuals with lower levels of perceived ability tend to prefer safer, uncontrollable options, whereas highly efficacious individuals favor controllable options, even in cases where more risk was involved [12]. Hence, investors with lower levels of online investment self-efficacy will likely find the traditional approach more appealing, as they will plausibly believe that a professional financial advisor would do a better job. On the other hand, individuals with higher levels of online investment self-efficacy will likely find the online investing environment more
alluring, as it offers the advantage of putting their perceived abilities to use. Thus, online investment self-efficacy should influence an investor's preference for traditional or online investing. Consistent with the generalizability framework and prior hypotheses, self-efficacies at higher levels in the hierarchy should be less predictive of preference. Hence, the following hypotheses are offered:
H3a: General self-efficacy will not have a
significant positive influence on preference for online investing.
H3b: Computer self-efficacy will not have a
significant positive influence on preference for online investing.
H3c: Online investment self-efficacy will have
a significant positive influence on preference for online investing.
4. Methodology
The purpose of this research was to develop and test an instrument that measures online investment self-efficacy (OISE) and demonstrate its ability to predict important phenomena related to online investing. To be considered a valid and reliable measure of the construct, the measure must demonstrate construct validity, which refers to the extent that the instrument measures the construct it intends to measure [8]. In order to establish construct validity, at a minimum, four criteria must be satisfied: 1) the measure must sufficiently capture the scope of the construct at the appropriate level of generalizability (content validity), 2) the items must measure the intended construct and no other construct (convergent validity), 3) the construct as a whole must differ from other related constructs (discriminant validity) and 4) the construct must be related to other constructs, as predicted by theory (nomological validity).
To fulfill these requirements, the instrument development process was divided into four phases. The first phase, item generation, served to create items for the online investment self-efficacy (OISE) instrument to ensure content validity. The second phase,
pretesting, assessed which of the items accurately
represented the OISE construct. The third phase,
validity and reliability assessment, established the
convergent and discriminant validity of the measure, as well as its reliability. The measure was compared to theoretically related measures in order to establish these properties. The fourth stage, hypotheses testing, empirically tested the veracity of the construct within
the context of the research (nomological) model to establish nomological validity. Each of these phases is described in the following subsections.
4.1. Item Generation
The purpose of the item generation phase was to ensure content validity, which refers to the extent to which the instrument contains representative items from a universal pool [17]. Items that accurately represent the underlying construct more reliably and precisely reflect the construct. However, the universal pool of items is virtually infinite, making the item generation process inherently difficult to verify [17]. To address this issue, domain experts were repeatedly consulted to evaluate the item pool until a consensus has been reached that the items sufficiently expressed the construct.
To our knowledge, no existing measure was available for the online investment self-efficacy construct. However, the IS literature provides useful suggestions regarding the development of self-efficacy measures, which can serve as a guide in creating items. Specifically, a five-step framework for operationalizing self-efficacy measures was adopted [14]. Care was taken to ensure each step was rigorously followed.
The literature suggests that a minimum of 5 or 6 items be included in the final instrument to ensure the set of indicators accurately measures the underlying construct [9]. It was highly unlikely that all items would load cleanly during the instrument development process. In this situation, items can either be 1) revised and administered to a separate sample, or 2) discarded altogether [6]. With this in mind, a sufficient number of items were generated to increase the probability that at least 5 or 6 items would be retained in the final scale.
To generate the preliminary list of items, one researcher conducted brainstorming sessions with two academicians with expertise in the area of online investing. The operational definition of the OISE construct and the five-step framework were given to each interviewee to ensure the items adhered to the specified definition and guidelines. The response format for each item incorporated both magnitude and strength ratings, resulting in an 11-place scale. Magnitude judgments assess an individual's perception of whether or not a particular task is attainable. A "I Could Not Do It" response (coded as 0) was incorporated to assess whether the respondent believed she or he could complete the task. Strength, which refers to one's level of confidence in completing a task, was captured via a scale anchored from 10% (Not At
All Confident) to 100% (Totally Confident) and coded from 1 to 10.
A draft instrument was created, which three independent academicians with expertise in psychometrics, computing, and investing were asked to review to identify any weaknesses. Reviewers were provided with an operational definition of the OISE construct and requested to comment on the instrument in general, as well as the individual items to eliminate ambiguities and potential misunderstandings. To enhance content validity, each reviewer was encouraged to identify irrelevant questions and offer suggestions for additional items that had been omitted. The draft instrument was modified accordingly, resulting in an initial pool of 25 items.
Thanks to comments from an anonymous reviewer on an earlier version of this paper, the initial pool of items was pared down based on Bandura’s position on self-efficacy measurement, “Investigators would do well to follow the dictum that the whole is greater than the sum of its parts” [3, p. 37]. Out of the initial item pool, 11 clearly characterized sub-skills (e.g. obtaining a price quote, analyzing price charts, finding news) rather than contextual beliefs pertaining to overall perceptions of one’s ability to invest online. Bandura adds, “A summation of decontextualized perceived efficacy of sub-skills would provide a misleading measure of operative capability” [3, p. 39]. Hence, these items were discarded, resulting in a preliminary set of 14 items that could be subjected to exploratory factor analysis in the pretesting phase (the actual instrument can be obtained by contacting the first author).
4.2. Pretesting
The preliminary instrument was administered to a pool of students in upper-division and graduate-level business courses at a large North American university. After eliminating cases with missing data, 127 usable responses were obtained. Responses were factor analyzed using principal components analysis as the factor extraction method. The Kaiser-Guttman Rule (Eigenvalues greater than 1) and the scree plot were utilized to determine the most appropriate component solution.
Referring to Table 1, the extraction revealed one dominant factor and one minor factor. Since the underlying factors were assumed to be correlated rather than orthogonal, promax rotation was utilized to yield a potentially more interpretable solution. Consistent with prior research, items that loaded at 0.707 or above on one factor [10] and less than 0.40 on any other factor [6] were utilized to identify acceptable items.
Table 1. Pretest factor loadings1 Factor2 Item Number 1 2 OISE13 0.718 0.476 OISE2 0.854 OISE33 0.743 0.547 OISE4 0.860 OISE5 0.913 OISE6 0.718 OISE7 0.844 OISE8 0.887 OISE9 0.925 OISE103 0.666 0.591 OISE11 0.939 OISE12 0.843 OISE13 0.911 OISE14 0.874 Eigenvalues 9.873 1.221 % of Variance 70.520 8.724 1
Kaiser-Meyer-Olkin Measure of Sampling Adequacy= 0.942
2For clarity, factor loadings < 0.40 are not shown 3Item discarded and not considered further
One researcher, one domain expert, and a psychometrician reviewed the results to determine the underlying factor structure. First, items 1, 3, and 10 cross-loaded on both factors, failing to meet the adopted thresholds. Second, a reasonable interpretation of the unique contribution of the minor factor could not be readily determined given the items’ phraseology. Third, for the second factor, the Eigenvalue value (1.221) narrowly met the Kaiser-Guttman criterion while the scree plot revealed an unmistakable elbow, signifying the factor’s negligibility. Based on this evidence, the panel agreed that items 1, 3, and 10 should be discarded and not considered further. The resulting instrument consisted of a pool of 11 items, which provided the basis for the validity and reliability assessment phase.
4.3. Validity and Reliability Assessment
Convergent validity refers to the state where the items measure their intended construct and no other construct [6], whereas discriminant validity refers to the state where the construct as a whole differs from other constructs [17]. Reliability refers to the state when a scale yields stable and consistent measures over time [17].
To establish a measure exhibits convergent and discriminant validity, the measure must be compared with other conceptually related constructs. In addition to the revised OISE instrument, existing computer self-efficacy (CSE) [7] and general self-self-efficacy (GSE) [5] measures were utilized. A preference measure was also included for the subsequent nomological validity phase. Following prior research [12], preference judgments were operationalized through the use of two vignettes, which provided descriptions of the
traditional and online investing approaches. After reading the vignettes, respondents were prompted to choose one and only one of the alternatives in which they would prefer to be involved and circled the desired option.
To avoid potential testing effects, the instruments were counterbalanced and administered to a pool of students in upper-division and graduate-level business courses at the same university. After removing subjects who participated in the pretesting phase and cases containing missing data, 314 usable responses were obtained. Overall, the sample averaged 23.4 years of age and consisted of 28% females. On 5-place scales anchored in 1 (Novice) and 5 (Expert), the typical subject reported above average computing expertise (M = 3.87) and limited investing expertise (M = 2.13). Roughly 69% of respondents indicated some prior exposure to online investing applications.
Table 2. Final factor loadings1
Factor2
Item Number
OISE CSE GSE Reliability3
OISE2 0.763 OISE4 0.876 OISE5 0.904 OISE6 0.821 OISE7 0.898 α = 0.975 OISE8 0.921 OISE9 0.924 OISE11 0.910 OISE12 0.849 OISE13 0.924 OISE14 0.912 CSE1 0.836 CSE2 0.804 CSE3 0.829 CSE4 0.820 α = 0.927 CSE5 0.831 CSE6 0.765 CSE7 0.745 CSE8 0.785 GSE1 0.772 GSE2 0.731 GSE3 0.789 GSE4 0.789 α = 0.905 GSE5 0.800 GSE6 0.782 GSE7 0.725 GSE8 0.761 Eigenvalues 10.223 5.267 3.604 % of Variance 37.863 19.507 13.347 1
Kaiser-Meyer-Olkin Measure of Sampling Adequacy= 0.933
2
For clarity, factor loadings < 0.40 are not shown
3Internal consistency reliability (Cronbach's
α)
Convergent validity was assessed thorough the examination of the factor loadings. Following the procedures outlined in pretesting, the items related to OISE, CSE, and GSE were subjected to principal components analysis and promax rotation. The literature suggests that CSE and GSE are
unidimensional scales [5,7]. Thus, three factors were expected to emerge. The analysis, however, revealed an unexpected factor pattern. Two CSE items (9 and 10) failed to converge on their intended construct, loading below the 0.707 threshold on CSE and cross-loading on a fourth, extraneous factor. Since these items did not exhibit convergent validity, a second round of analysis was undertaken without them. As depicted in Table 2, convergent validity was established, as the remaining items loaded strongly on their associated factors and not on any other factor. Thus, the two problematic items were discarded and not considered further.
Discriminant validity was assessed by comparing the average variance extracted (AVE) associated with each construct to the correlations among constructs [16]. Table 3 shows the results of the discriminant validity analysis. Diagonal elements represent the square root of the AVE, whereas the off-diagonal elements represent the correlations among constructs. In order to claim discriminant validity, the diagonal elements should be larger than any other corresponding row or column entry [16]. According to the calculations, each construct sufficiently differed from the other constructs and, therefore, the measures demonstrated discriminant validity.
Reliability was assessed in two ways: 1) internal consistency reliability (Cronbach's α), and 2) AVE. Referring to Table 2, internal consistency reliability (α) substantially exceeded the generally agreed upon lower limit of 0.70 [15]. Since α tends to be susceptible to the number of items in a scale, AVE was utilized as a more stringent test of reliability. As shown in Table 3, all AVE's surpassed the commonly accepted 0.50 level [16], suggesting that the items captured over 50% of the variance in their respective constructs. Both α and AVE calculations confirmed the measures' reliability. In sum, the final solution demonstrated admirable levels of convergent and discriminant validity, as well as reliability.
Table 3. Correlations and AVE
Construct1
Construct AVE OISE CSE GSE
OISE 0.780 0.883
CSE 0.644 0.303*** 0.802
GSE 0.591 0.148** 0.272*** 0.769
1Diagonal elements (in bold) represent the square root of the average variance
extracted (AVE). Off-diagonal elements represent the correlations among constructs.
*p < 0.05, **p < 0.01, ***p < 0.001.
4.4. Nomological Validity
Once the psychometric merits of the OISE measure were confirmed, nomological validity could be ascertained. Nomological validity refers the state
where the construct is related to other constructs as predicted by theory [8]. The research model comprised a comprehensive nomological network under which this criterion could be examined.
Hierarchical regression techniques were utilized to formally test the hypotheses. This approach involves a series of regressions, working "backward" in the model. Since preference, OISE, and CSE represent dependent variables at specific stages in the model, the hypothesized direct paths and mediating processes leading the dependent variable of interest are evaluated in isolation. Thus, the hypotheses related to the dependent variable preference were tested first, followed online investment self-efficacy, and, finally, computer self-efficacy.
The first stage involved testing hypotheses H3a,
H3b, and H3c, which examined the influence of GSE,
CSE, and OISE on preference respectively. Since the preference variable was dichotomous in nature (coded as 0 = traditional, 1 = online), logit regression was the most appropriate analysis technique. Furthermore, in total the hypotheses suggest that OISE fully mediates the effects of GSE and CSE on preference. To test the direct effects or GSE, CSE, and OISE on preference, as well as account for OISE mediation, the independent variables GSE and CSE were entered into the logit regression model first, followed by OISE. Preference was used as the dependent variable. The Pseudo R2 change (∆R2
) was scrutinized to identify the mediating role of OISE. The results for the first stage of analysis indicated the GSE was not a significant predictor of preference (z = 0.74, p = 0.461), whereas CSE was significant (z = 2.34, p = 0.019). The second stage revealed a significant improvement in predictive power (∆R2
= 0.062, Pseudo R2 = 0.079, ∆F = 21.05, p < 0.001). GSE continued to be non-significant (z = 0.43, p = 0.668), CSE became non-significant (z = 1.19, p = 0.235), and OISE proved to be a significant predictor (z = 4.81, p < 0.001). Given these results, H3a, H3b, and
H3c were supported and, as expected, OISE fostered
preference for the online investing approach (as indicated by the significant and positive z-score) and fully mediated the effects of GSE and CSE.
The next stage involved testing hypotheses H2a
and H2b, which examined the antecedents of OISE,
namely GSE and CSE respectively. In total, the hypotheses predicted that CSE would fully mediate the effect of GSE on OISE. Since all variables were continuous in nature, linear regression was the most appropriate analysis technique. To account for the mediating role of CSE, GSE was entered into the regression model first, followed by CSE. OISE was used as the dependent variable. The results for the first stage indicated that GSE was a significant predictor of
OISE (β = 0.148, t = 2.64, p = 0.009). The second stage revealed a significant improvement in predictive power (∆R2
= 0.074, R2 = 0.096, ∆F = 25.63, p < 0.001). The effect of GSE became non-significant (β = 0.071, t = 1.27, p = 0.207) while CSE proved significant (β = 0.284, t = 5.06, p < 0.001). Thus, the results supported H2aand H2b, suggesting that CSE was
a significant predictor of OISE and fully mediated the effect of GSE.
The final stage involved testing H1, which
examined the relationship between GSE and CSE. This hypothesis was tested via linear regression using GSE and CSE as the independent and dependent variables respectively. The results revealed that GSE was a significant predictor of CSE (β = 0.272, t = 4.992, p < 0.001), supporting hypothesis H1.
5. Discussion and Conclusion
The purpose of this endeavor was to build and test a measure for online investing self-efficacy and demonstrate its power to predict an important online investing phenomenon – individual preference for the traditional or online approach to investing. Based on the results, these objectives were achieved successfully. The measure exhibited admirable psychometric properties and provided general support the proposed research model based on the perspective of Social Cognitive Theory. In the process, this research filled notable voids in the self-efficacy generalizability framework by empirically establishing relationships between general efficacy and self-efficacies in the realm of computing at both the domain and task levels.
In addition to achieving its objectives, this research revealed an interesting insight into the nature of the CSE measure. The CSE measure utilized [7] has been criticized for noncompliance with respect to the formal definition of self-efficacy and, therefore, appears to lack face validity [14]. Evidence emerging from the validity analyses supports this contention, as two items were removed due to their inability to load cleanly on the construct. The findings suggest that researchers should exercise caution when utilizing this measure by examining its psychometric properties within the context of their theoretical frameworks.
Despite making valuables contribution to our knowledge, it must be acknowledge that this research was limited in certain respects. First, the data analyses utilized common factor analysis and hierarchical regression techniques, which have been criticized for their assumption of perfect measurement. Structural equation modeling (SEM) may provide a more powerful technique to facilitate measurement
precision. SEM can also model complex relationships in a set of hierarchical structural equations to provide a comprehensive picture of the model. By empirically purging measurement error and simultaneously estimating structural relationships, clearer conclusions may have been drawn. Given these arguments, additional research is clearly needed to address these issues.
It must be noted that this endeavor utilized student respondents, who may not be entirely representative of real-world online investors. However, with the exception of age, the sample characteristics compared favorably to the typical online investor [4]. Perhaps more importantly, these subjects are the investors of the future, which allows educated inferences to be made concerning the manner in which they are likely to invest given the opportunity. Regardless, the sample provided initial support for the theoretical model. There is no empirical evidence to date that suggests that students will react differently from real-world online investors, although differences will likely emerge in terms of financial wherewithal, risk tolerance, and investing experience. Ultimately, however, the issue of generalizability is best addressed through replication in different contexts using complementary samples to identify the boundary conditions of the measure and theoretical model [9].
Going forward, the robust baseline theory developed herein could be leveraged to examine more specific phenomena, teasing out important variables in richer contexts. Several potential applications of the OISE measure exist, some of which are mentioned here. Research suggests that investor overconfidence fosters hyperactive online trading and, ultimately, poor performance due to mounting transaction costs [4]. Self-efficacy has been linked to expectations of future outcomes [11]. Thus, OISE may play a central role in the formation of online investor overconfidence. Similarly, the construct could help explain phenomena such as herding behavior, as verbal persuasion and psychological states can lead to heightened self-efficacy beliefs [3]. The reciprocal nature of SCT could also enable the examination of OISE and its impact on online investing behavior over time. A longitudinal assessment could offer new insights into online investor psychology and behavior as they evolve in response to changing market conditions.
6. References
[1] Agarwal, R., Sambamurthy, V. and Stair, R.M. (2000), "Research Report: The Evolving Relationship Between General and Specific Computer Self-Efficacy - An Empirical Assessment," Information Systems Research, 11:4, 418-430.
[2] Bandura, A. (1986), Social Foundation of Thought and
Action: A Social Cognitive Theory, Prentice Hall: Englewood
Cliffs.
[3] Bandura, A. (1997), Self-Efficacy: The Exercise of
Control, W.H. Freeman and Company: New York.
[4] Barber, M.B. and Odean, T. (2000), "Trading is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors," Journal of Finance, 55:2, 773-806.
[5] Chen, G., Gully, S.M. and Eden, D. (2001), "Validation of a New General Self-Efficacy Scale," Organizational
Research Methods, 4:1, 62-83.
[6] Chin, W.W., Gopal, A., and Salisbury, W.D. (1997), "Advancing the Theory of Adaptive Structuration: The Development of a Scale to Measure Faithfulness of Appropriation," Information Systems Research, 8:4, 342-367. [7] Compeau, D.R. and Higgins, C.A. (1995), "Computer Self-Efficacy: Development of a Measure and Initial Test,"
MIS Quarterly, 19:2, 189-211.
[8] Cook, T.D. and Campbell, D.T. (1979),
Quasi-Experimentation: Design and Analysis Issues for Field Settings, Rand McNally: Chicago.
[9] Dennis, A.R. and Valacich, J.S. (2001), "Conducting Experimental Research in Information Systems,"
Communications of the AIS, 7:5, 1-41.
[10] Gefen, D., Straub, D.W., and Boudreau, M.C. (2000), "Structural Equation Modeling and Regression: Guidelines for Research Practice," Communications of the AIS, 4:7.
[11] Johnson, R.D. and Marakas, G.M. (2000), "Research Report: The Role of Behavioral Modeling in Computer Skill Acquisition: Toward Refinement of the Model," Information
Systems Research, 11:4, 402-417.
[12] Klein, W.M. and Kunda, Z. (1994), "Exaggerated Self-Assessments and the Preference for Controllable Risks,"
Organizational Behavior and Human Decision Processes,
59, 410-427.
[13] Looney, C.A. and Chatterjee, D. (2002), "Web-Enabled Transformation of the Brokerage Industry," Communications
of the ACM, 45:8, 75-81.
[14] Marakas, G.M., Yi, M.Y., and Johnson, R.D. (1998), "The Multilevel and Multifaceted Character of Computer Self-Efficacy: Toward a Clarification of the Construct and an Integrated Framework for Research," Information Systems
Research, 9:2, 126-162.
[15] Nunnally, J.C. and Bernstein, I.H. (1994), Psychometric
Theory, 3rd edition, McGraw Hill: New York.
[16] Staples, D.S., Hulland, J.S., and Higgins, C.A. (1999), "A Self-Efficacy Theory Explanation for the Management of Remote Workers in Virtual Organizations," Organization
Science, 10:6, 758-776.
[17] Straub, D.W. (1989), "Validating Instruments in MIS Research," MIS Quarterly, 13:2, 147-166.
[18] Taylor, S. and Todd, P.A. (1995), "Understanding Information Technology Usage: A Test of Competing Models," Information Systems Research, 6:2, 144-176.