Four hundred and forty college students voluntarily participated in our survey. The response rate is 17.6 per cent. One hundred and thirty-eight respondents (31.4 per cent) are male and 302 female (68.6 per cent). The mean age of the sample is 23.9. An overwhelming majority of them are Whites (90.9 per cent) and minority students make up 9.1 per cent of the sample.
As for the typical daily uses of their cell phones, respondents spent 61.5 min talking on their cell phones on average with the standard deviation of 72.4 (the median=30 min and the mode=30 min). On a typical day, the participants sent 54 text messages on average with the standard deviation of 84.6 (the median=24 and the mode=50). Their typical daily uses of cell phones for talking and text messaging varied considerably, ranging from 0 to 540 min of talk time and from 0 to 650 text messages.
Table 1 presents Cronbach's coefficients (α) of all adopted and adapted scales and the results of exploratory factor analyses (principle axis factoring with varimax rotation). A liberal minimum requirement for scale reliability is 0.60,37, 38 while some scholars recommended a stricter minimum requirement of 0.70.39 So, the performance of three scales is very satisfactory considering their high Cronbach's coefficients and high percentage of variance explained. However, one scale proved to be not reliable.
The weak performance of the perceived risks scale can be explained by the fact that most American cell phone users have recognized the monetary costs associated with passing along mobile messages but not all of them are concerned about the loss of time. Therefore, the question about recipients’ monetary costs was dropped from the perceived cost scale.
Pearson's correlation analysis was conducted to test the correlational Hypothesis 3b. As predicted, American young consumers’ subjective norm was positively associated with perceived control (r=0.37, P<0.01).
Structural equation modeling method with Amos 18 was used to test our proposed research models and directional hypotheses. The fitness indexes of eight tested models are shown in Table 2. Figures 3, 4, 5, 6, 7 and 8 present the standardized path estimates of all variables in two proposed TPB models and four re-specified models. Significant results of six χ2 statistics imply that those six models were not acceptable. However, the χ2 statistic alone cannot successfully assess the fitness of models estimated with large samples as the likelihood ratio test is very sensitive to sample size and it also assumes that the model fits perfectly in the population. Thus, other fitness indexes were developed to address this problem.40 The root mean square error of approximation (RMSEA) of a good model should be equal to or smaller than the recommended cutoff value41 of 0.06 while its goodness of fit index (GFI) should reach the conventional acceptable level of 0.90 if it cannot meet the stricter standard of 0.95 suggested by recent scholars.42 Its Tucker–Lewis index (TLI) or non-normed fit index (NNFI) should be >0.95, the widely accepted cutoff for a good model fit.41, 42 By convention, its comparative fit index (CFI) should be ⩾0.90 to accept the model, indicating that 90 per cent of the covariation in the data can be reproduced by the given model. On the basis of these criteria, the proposed TPB model 1a and 2a have achieved satisfactory fit even though their RMSEA's are >0.06. A big RMSEA suggests that a mediator should be identified and so the TPB model was re-specified in which mobile viral marketing intent served as a predictor and mediator. As a result, the re-specified TPB model 1b and 2b have accomplished perfect fit.
The path estimates from viral marketing attitude to mobile viral marketing intent and from intent to behavior in Figures 3, 4, 5, 6, 7 and 8 all strongly supported Hypothesis 1 and Hypothesis 2. Young American consumers’ viral marketing attitude predicted their intent to pass along electronic messages, which led to their actual behavior of forwarding mobile messages. Their viral marketing intent mediated the effect of attitude on behavior.
The proposed model 3 and 4 of mobile viral marketing did not fit the data satisfactorily at first. However, they achieved a pretty good fit, respectively, after re-specifying viral marketing intent as a mediator and correlating those error items as suggested by modification indexes of Amos. Figures 7 and 8 present the standardized path estimates of two revised models of mobile viral marketing.
The significant estimated path coefficients from subjective norm to viral marketing attitude in Figures 7 and 8 supported Hypothesis 3a. Hypothesis 4a and Hypothesis 7a were also substantiated by significant path estimates of behavioral control and perceived cost leading to attitude toward viral marketing in Figures 7 and 8. Hypothesis 5a and Hypothesis 6a were rejected as perceived utility and EOU did not predict viral marketing attitude. Two significant predictors of young American consumers’ mobile viral marketing intent consistently emerged in Figures 7 and 8: viral marketing attitude and perceived utility. So, Hypothesis 5b was supported. Subjective norm did not predict American young consumers’ intent to pass along entertaining and useful messages in two final models. Therefore, Hypothesis 3c was rejected. Hypothesis 4b, Hypothesis 6b and Hypothesis 7b, were not supported as there were no significant path coefficients connecting behavioral control, EOU and perceived cost with young American consumers’ mobile viral marketing intent in Figures 7 and 8. The results of hypothesis testing were summarized more clearly in Table 3.
Four backward regression procedures were conducted to explore what demographic, psychographic and behavioral factors influence American young consumers’ mobile viral marketing attitude, intent and behavior.
The following five variables survived eight iterations to predict American young consumers’ attitude toward viral marketing, listed in order of significance: subjective norm (β=0.421, t=10.01, P<0.001), perceived cost (β=−0.220, t=−5.79, P<0.001), behavioral control (β=0.168, t=4.04, P<0.001), perceived utility (β=0.08, t=1.89, P=0.059) and text messaging (β=0.062, t=1.65, P=0.100). These variables were responsible for 41.4 per cent of the variance in their attitude toward viral marketing.
To determine what demographic, psychographic and behavioral factors are significant predictors of American young consumers’ behavioral intention to engage in viral marketing, two backward multiple regressions were operated on the following 13 independent variables: gender, race, age, family annual income, personal monthly income, cell phone calling, text messaging, attitude toward viral marketing, subjective norm, perceived behavioral control, perceived utility, EOU and perceived costs. Three significant predictors of their intent to pass along entertaining electronic messages were retained after 11 iterations in order of their importance: attitude toward viral marketing (β=0.346, t=7.57, P<0.001), perceived utility (β=0.198, t=4.34, P<0.001) and telephoning (β=−0.086, t=−2.00, P=0.046). Over 20 per cent of the variance in American young consumers’ intent to forward entertaining viral messages (R2=0.204) can be accounted for by these three factors. Another backward regression was run against American young consumers’ intent to pass along useful electronic messages to test these hypotheses. Nine iterations yielded the following five significant predictors: perceived utility (β=0.203, t=4.171, P<0.001), attitude toward viral marketing (β=0.159, t=2.97, P=0.003), subjective norm (β=0.120, t=2.176, P=0.030), income (β=0.089, t=2.026, P=0.043) and telephoning (β=0.084, t=1.905, P=0.057). They explained 17.2 per cent of the variance in young American consumers’ intent to pass along useful messages (R2=0.172).
The last backward regression revealed seven significant predictors of American young consumers’ actual mobile viral marketing behavior after nine iterations: attitude toward viral marketing (β=0.239, t=4.922, P<0.001), intent to forward entertaining electronic messages (β=0.122, t=2.446, P=0.015), telephoning (β=0.111, t=2.47, P=0.014), text messaging (β=0.105, t=2.384, P=0.018), intent to forward useful electronic messages (β=0.097, t=2.014, P=0.045), race (β=−0.095, t=−2.189, P=0.029) and gender (β=0.079, t=1.821, P=0.069). These seven factors accounted for 19.5 per cent of the variance in American young consumers’ actual behavior of forwarding mobile messages (R2=0.195).
Generally, the results of four backward regressions confirmed that of structural equation modeling procedures.