*4.2. Measurement*

The survey was designed using a multi-item approach. All variables were carried out by a five-point Likert-scale, ranging from strongly disagree (1) to strongly agree (5). Items were borrowed from previous literature and modified for the context of this study.

The questionnaire consisted of three parts:


The whole items are shown in Table 3.

> **Table 3.** Summary of measurement scales constructs (in the table, short video refers to health-related short videos).


#### **5. Data Analysis and Results**

The research model was tested using AMOS 21, a structural modeling technique that is well suited for estimating parameters and theoretical models. In Section 5.1, we use confirmatory factor analysis (CFA) to validate the reliability and validity of the measurement model. In Section 5.2, we perform maximum likelihood estimation (MLE) and the bootstrapping method to examine the structural model. In Section 5.3, we measure the mediating effects of PU and AT in the research model. In Section 5.4, we summarize the conclusion of the research.

Before measuring the model, we examined whether common method bias was a concern in this study with two tests of common method variance (CMV) [44]. First, an exploratory factor analysis of all items extracted eight factors, which explained 73.72% of all the variance, with no single factor accounting for significant loadings (at the *p* < 0.10 level) for all items. Second, single method-factor approaches indicated that there was no significant difference among the method factors (ΔCMIN/DF = −0.032, ΔGFI = 0.009, ΔAGFI = 0.001, ΔIFI = 0.003, ΔRMSEA = −0.002). Thus, we concluded that CMV was not a concern in this data set.

#### *5.1. Measurement Model Evaluation*

The measurement model evaluation mainly included three parts, namely reliability analysis, convergen<sup>t</sup> validity, and discriminant validity.

Reliability is usually examined by using internal consistency reliability and composite reliability. As shown in Table 4, the coefficient alpha for the eight variables scored in the range from 0.800 to 0.907 (more than 0.70 [45]). The composite reliability (CR) of all constructs was above 0.70 [46]. The reliability of the model was achieved.

Convergent validity is usually examined by using composite reliability (CR) and average variance extracted (AVE). As shown in Table 4, all items load significantly on their respective constructs and none of the loadings is below the cutoff value of 0.70. AVE of each variable was more than 0.50 [47]. Thus, convergen<sup>t</sup> validity was achieved.

Discriminant validity is analyzed to examine whether a measurement is not a reflection of any other measurement or not. As shown in Table 5, the square root of AVE for each variable is greater than the other correlation coefficients. The discriminant validity of the model is achieved.


**Table 4.** Statistics of construct items.


**Table 4.** *Cont.*

**Table 5.** Correlations between constructs (AVE and squared correlations).


Note: The numbers in the diagonal row are square roots of the average variance extracted.

#### *5.2. Structural Model Evaluation*

The research model was tested by using AMOS 23, and a bootstrapping resampling procedure (2000 samples) was used to ensure the solidity.

The resulting indices indicated a good model fit (x2/df = 1.318; GFI = 0.903; RMSEA = 0.035; NFI = 0.909; CFI = 0.976; TLI = 0.972; RMR = 0.071; SRMR = 0.058).

Figure 3 shows that social interaction (SI), informativeness (IN), and relevance (RE) of health-related short-video ads were significant and positive predictors of perceive usefulness of ads (β = 0.23, *p* < 0.01; β = 0.22, *p* < 0.01; β = 0.20, *p* < 0.01). These findings support H1, H3, and H4, respectively, and this supports previous studies [12,19]. Similarly, intrusiveness (IR) of health-related short-video ads was a significant and negative predictor of the perceive usefulness of ads (β = −0.21, *p* < 0.01), and H2 is validated. This is consistent with previous research [2,15]. These four factors (SI, IR, IN, and RE) explained 29% of the variance in the perceived usefulness of health-related short-video ads. Additionally, comparing these four coefficients, we found social interaction was more important to users' perceived usefulness of ads. Users' perceived usefulness of health-related short-video ads positively affected their attitudes toward ads (β = 0.59, *p* < 0.001) and purchase intention (β = 0.21, *p* < 0.05); these results validate H5 and H6. Users' perceived ease of use of health-related short-video ads was a significant and positive predictor of perceived usefulness of ads (β = 0.27, *p* < 0.001), and H7 is validated. However, users' perceived ease of use of health-related short-video ads could not

significantly predict users' attitudes toward ads. H8 is not validated and this is not consistent with Lin and Kim' research, as they found users' perceived ease of use of social media ads significantly predict users' attitudes [2]. This enlightens us that perceived ease of use may not be an important factor in health-related short-video ads. Therefore, only users' perceived usefulness of ads explained 37% of the variance in attitudes toward ads. Finally, users' attitudes of health-related short-video ads were found to positively predict their purchase intentions (β = 0.23, *p* < 0.05), and H9 is validated. Users' perceived usefulness of ads and attitudes toward ads accounted for 33% of the variance in purchase intentions.

**Figure 3.** Results of structural model evaluation.

#### *5.3. Mediating E*ff*ects Analysis*

Mediation analyses were tested using the Preacher and Hayes' bootstrapping (2000 bootstrap samples) analysis with the PROCESS macro [48]. We focused on two mediate variables (perceived usefulness and ad attitudes) to find the impact path of user responses to health-related short-video advertisements because users' perceived ease of use of health-related short-video ads cannot significantly predict users' attitudes toward ads. The results are shown in Table 6.

The mediating path (Social Interaction→PU→AT) tested the specific indirect effect of social interaction of ads on ad attitudes through perceived usefulness of ads. As both the upper and lower limits of the confidence interval (CI) were above zero (0.014, 0.166), this indirect effect can be interpreted as significantly positive. Similarly, the direct effect of the perceived usefulness of ads can be interpreted as significantly positive. Hence, the perceived usefulness of ads served as a partial mediating variable between social interaction and ad attitudes. That means those who perceived the sociality of health-related short-video ads believed that the ad was more useful, and this increased usefulness enhances the viewers' positive attitudes toward the ad.

The mediating path (Intrusiveness→PU→AT) tested the specific indirect effect of intrusiveness of ads on ad attitudes through perceived usefulness of ads. As both the upper and lower limits of the confidence interval (CI) were above zero (−0.134, −0.018), this indirect effect can be interpreted as significantly positive. The perceived usefulness of ads served as a partial mediating variable between intrusiveness and ad attitudes. That means those who perceived intrusiveness of health-related short-video ads believed that the ad was less useful, and this decreased usefulness enhances the viewers' negative attitudes toward the ad.


**Table 6.** Mediating effects analysis.

The mediating path (Informativeness→PU→AT) tested the specific indirect effect of informativeness of ads on ad attitudes through perceived usefulness of ads. Perceived usefulness of ads served as a partial mediating variable between informativeness and ad attitudes (CI: 0.011, 0.164). That means those who perceived informativeness of health-related short-video ads believed that the ad was more useful, and this increased usefulness enhances the viewers' positive attitudes toward the ad.

The mediating path (PEOU→PU→AT) tested the specific indirect effect of perceived ease of use of ads on ad attitudes through perceived usefulness of ads. Perceived usefulness of ads served as a partial mediating variable between perceived ease of use and ad attitudes (CI: 0.024, 0.188). That means those who perceived ease of use of health-related short-video ads believed that the ad was more useful, and this increased usefulness enhances the viewers' positive attitudes toward the ad.

Similarly, the perceived usefulness of ads served as a partial mediating variable between perceived ease of use and purchase intentions (CI: 0.018, 0.195). That means those who perceived ease of use of health-related short-video ads believed that the ad was more useful, and this increased usefulness enhances the viewers' purchase intentions.

Finally, the mediating path (PU→AT→PI) tested the specific indirect effect of perceived usefulness of ads on purchase intention through ad attitudes. The indirect effect was significant (CI: 0.026, 0.368) while the direct effect was not significant (CI: −0.007, 0.511). Hence, ad attitudes served as a full mediating variable between perceived usefulness and purchase intention. That means users' attitude toward health-related short-video ads was a key path to the effectiveness of ads.

In summary, there were four impact paths of user responses to health-related short-video advertisements. First, social interaction had an indirect effect on purchase intentions through perceived usefulness and ad attitudes in serial (SI→PU→AT→PI). Second, intrusiveness had an indirect effect on purchase intentions through perceived usefulness and ad attitudes in serial (IR→PU→AT→PI). Third, informativeness had an indirect effect on purchase intentions through perceived usefulness and ad attitudes in serial (IN→PU→AT→PI). Fourth, perceived ease of use had an indirect effect on purchase intentions through perceived usefulness and ad attitudes in serial (PE→PU→AT→PI).

#### *5.4. Summary of Hypothesis Validation*

By combing the results described in Section 5.1–Section 5.3, we present the summary of the hypothesis validation in Table 7.



The validation-result column in Table 7 shows the final validation results of each hypothesis, from which we can see that H1-H4 are well established. These hypotheses correspond with the research questions RQ1 and RQ2 presented in Table 1. This implies that five qualities (social interaction, intrusiveness, informativeness, relevance, and perceived ease of use) of health-related short-video ads help enhance the perceived usefulness of ads. Among all the factors, according to different coefficients, social interaction of health-related short-video ads made a more valuable contribution to advertising usefulness. Thus, advertisers and marketers should not only focus on social interaction, intrusiveness, informativeness, relevance, and the perceived ease of use of health-related short-video ads, but also pay more attention to social interaction.

Regarding hypotheses H5, H7, and H8, which aim to answer RQ3, only the perceived usefulness of health-related short-video ads helped produce positive user attitudes. H9 answers RQ4, finding that positive user attitudes significantly promotef purchase intentions. Combined with the analysis of the mediating effects in Section 5.3, we found four impact paths of user responses to health-related short-video advertisements (SI→PU→AT→PI; IR→PU→AT→PI; IN→PU→AT→PI; PE→PU→AT→PI). This answers RQ5. Thus, advertisers and marketers may maximize resource utilization through the impact paths.
