A social cognitive theory of Internet uses and gratifications: toward a new model of media attendance.

Relationships among deficient self-regulation, Internet self-efficacy, habit strength, and Internet usage were explored previously in research on so-called "Internet addictions" among college students (LaRose et al., 2003) and can be summarized as follows:

H1: Internet self-efficacy will be directly related to Internet usage.

H2: Internet habit strength will be directly related to Internet usage.

H3: Deficient Internet self-regulation will be directly related to Internet usage.

H4: Deficient Internet self-regulation will be directly related to Internet habit strength.

H5: Internet self-efficacy will be directly related to Internet habit strength.

The causal relationship of prior Internet experience to Internet self-efficacy has also been verified in college student populations (Eastin & LaRose, 2000). Self-efficacy is conceived as the product of progressive mastery of behavior that increases with experience. If habit is simply a recurring behavior pattern, then the amount of prior experience with the Internet should be directly related to habit strength. Theoretically, repetition fosters a growing inattentiveness to behavior, undermining self-monitoring. Triandis (1980) pointed out that the best predictor of behavior is past behavior. Indeed, this truism may be largely responsible for the scant attention paid to habits in media research, since it has the ring of a tautology: One measure of behavior predicts another measure of behavior (and so what?). However, we propose that the impact of prior experience on current behavior is explainable entirely through the causal paths among socio-cognitive constructs.

H6: Prior Internet experience will be directly related to Internet self-efficacy.

H7: Prior Internet experience will be directly related to Internet habit strength.

To complete the model of Internet usage, the role of outcome expectations (a.k.a uses and gratifications) also must be specified. LaRose et al. (2003) examined a single type of outcome expectation, self-reactive outcomes, which uses and gratifications researchers would perhaps recognize as "pass time" gratifications. They found that self-reactive outcome expectations were causally related to Internet usage and also acted on Internet usage through deficient self-regulation. Self-reactive outcome expectations were themselves preceded by Internet self-efficacy and followed by habit strength.

However, other types of outcome expectations should also explain Internet usage. LaRose et al. (2001) found that the gratifications of the Internet, reconceptualized as outcome expectations reflecting other incentive categories recognized by SCT, were positively related to Internet usage. However, these relationships were not arrayed in an overall causal model and self-reactive and activity outcomes were combined to attain satisfactory reliability. Thus, the self-reactive and activity constructs will be separated to match the conceptual distinction between these two incentive categories.

Self-efficacy should precede each of the the various types of outcomes and habit strength should follow them. Users need to learn how to successfully obtain social, status, activity, novelty, and monetary gratifications as much as self-reactive ones. But once they achieve satisfactory means for attaining those outcomes, they should become increasingly inattentive to specific behaviors that support them.

Past research expended a great deal of effort distinguishing gratification dimensions, but from the SCT perspective expected outcomes are a unitary construct, representing the mechanism of learning through experience. Thus, we represent gratifications, now understood to be different types of expected outcomes grouped into six incentive categories, as first-order concepts that are part of a second-order construct, outcome expectations.

H8: Expected a) activity, b) social, c) status, d) novel, e) self-reactive, and f) monetary Internet outcomes will be directly related to Internet usage.

H9: Internet serf-efficacy will be directly related to expected Internet outcomes.

H10: Expected Internet outcomes will be directly related to Internet habit strength.

An important exception is noted for the relationship between self-reactive outcomes and deficient self-regulation. Self-reactive outcomes bear a unique relationship to deficient self-regulation. The use of the media to adjust internal states should be the main type of incentive susceptible to triggering the spiral of excessive usage and dysphoria thought to lead to problematic media usage. Thus,

H11: Self-reactive outcomes of Internet usage will be positively related to deficient Internet self-regulation.

These hypothesized relationships are represented in the path diagram shown in Figure 1. The final path analysis (see Results, below) was identical except for one path that was proposed but not statistically significant, another that narrowly missed significance (both indicated by light dotted lines) and another that was not initially proposed but uncovered during analysis (indicated by a heavy dotted line), in the interest of space only the final path diagram is shown. For the sake of clarity the word "Internet" has been omitted from the labels in figures and tables.

[FIGURE 1 OMITTED]

Research Methods

Procedure

To obtain a diverse sample of the general population of adult Internet users, respondents were recruited by mail from two Midwestern communities to complete an online survey in April and May of 2002. Both communities included a major university and surrounding counties. A commercial mailing list vendor provided a random sample of household addresses in the designated communities. The initial mailing included a letter advising respondents of the purpose of the study and their rights as human subjects. Half the letters requested that the survey be filled out by a male head of household and the other half by a female head of household, if such a person were available. Also included in the envelope was a nominal cash incentive and a postcard with the URL and a respondent ID for the survey. Internet users were instructed to use the card and ID number the next time they went on the Internet. Non-users were instructed to indicate their gender and year of birth and return the card by mail so that response rates could be calculated and the results compared to U.S. Census data.

Respondents

Of the 1100 solicitations sent, 170 (15%) bad addresses were returned; leaving a total usable sample of 930. A total of 331 responded to the solicitation. One hundred and seventy-two Internet users completed the survey and 159 returned the non-Internet user postcard (36% total response rate). Recent research assessing response rates (Yun & Trumbo, 2000) indicates that the present rates were consistent with methods employing Web surveys. A rule of thumb for structural equation modeling, and also for regression analysis, is that there be 10 respondents for each link/ relationship in the model. The proposed model has 16 links so the sample size was deemed adequate. There were no response difference by city and thus data were collapsed. As a total sample (N = 331) participants were 55% male and 45% female. The general population in the counties surveyed was 50% female (U.S. Census, 2002). Six percent of the participants were between the ages of 18-24 (census population = 17%), 48% were between the ages of 25-44 (census population = 30%) 34% were between 45-65 years old (census population = 40%), and finally, 13% were over the age of 65 (census population = 14%). The respondents were thus a somewhat biased sample of the respective populations from which they were drawn. However, a diverse sample of adult respondents was obtained and therefore this sample was deemed suitable for the purpose of this study which was to examine relationships between variables.

The non-Internet users (N = 159) were 48% male and 52% female and their mean age was 52 years old. Given current estimates of Internet penetration (54%, NTIA, 2002) we estimated that respondents at 504 (out of 938) of the valid addresses had access to the Internet, and thus, could have completed the online survey. Therefore, we estimate that the 172 people who completed the survey represent an Internet user response rate of 34%. Of those, 41% were female and 58% were male (with 1% not indicating their gender) with an average age of 42 years old. Eighty-nine percent were Caucasian, 5% were African American, 2% were Latino and the remaining 4% were Asian, Pacific Islander, Native American, or other. Forty-two percent of the sample had average household incomes under $50,000; the remaining 58% had incomes greater than $50,000. Educationally, participants ranged between 9-22 years beyond kindergarten (Mean = 16, S.D. = 2.61). Six respondents were removed from the sample for incomplete data, yielding a final sample of 167. A common rule of thumb in structural equation modeling is to have 10 cases for each link in the model. The proposed model had 16 links and so the size of the sample was deemed adequate.