CHAPTER 4: CONTENT ANALYSIS RESULTS
4.4 Statistical Analysis of the Relationship between Credibility and Credibility Factors
4.4.1 Correlation Analysis
Studying the correlation between credibility and underlying factors revealed substantial insight. The correlation coefficients and p-values are summarized in Table 13. The credibility factors in Table 13 are listed in descending order of absolute value to correlation coefficient, and the leftmost column explains the social media type to which a corresponding credibility factor is applicable. Plausibility showed the largest positive correlation (0.75, p < .001) with the
credibility, while promotion and negativeness show the largest negative correlation (-0.32, p < .001). In addition to the plausibility, objectiveness, comprehensiveness, and relevance showed a moderate positive relationship with credibility. All other underlying factors showed a weak positive relationship with credibility. Except for reference, all underlying factors of credibility
showed statistically significant differences at α = 0.05 level. In other words, the correlation coefficients of underlying factors except reference are significantly different from 0.
Table 13. Correlation between Credibility and Underlying Factors (Yahoo! Answers) Social Media Type Credibility Factor Correlation p-value
Unique (Yahoo! Answers)
Plausibility 0.75 0.00E+00***
Relevance 0.57 0.00E+00***
Sympathy 0.21 2.69E-06***
First-hand experience 0.12 6.74E-03***
Both (Yahoo! Answers and Yelp) Objectiveness 0.61 0.00E+00*** Comprehensiveness 0.58 0.00E+00*** Negative -0.32 2.90E-13*** Promotion -0.32 1.95E-13*** Expertise 0.28 2.50E-10*** Honesty 0.19 2.64E-05*** Presentation 0.17 1.92E-04*** Positiveness 0.16 3.18E-04*** Currency 0.12 9.11E-03*** Credence 0.11 1.83E-02* Reference 0.08 6.40E-02
* significant at p < 0.05; ** significant at p < 0.005; *** significant at p < 0.001 The correlations between pairs of credibility factors could provide several important insights. First, this analysis could help to understand the pattern of frequently co-occurring credibility factors. Second, those patterns of co-occurrence could reveal the existence of potential interactions between credibility factors. Most of the correlations calculated between credibility
factors were low and not statistically significant, indicating weak co-occurrence patterns. However, correlations between several factors showed a strong or moderate relationship. Thus, the top 10 correlation pairs were reported in Table 14.
Table 14. Correlations between Credibility Factors (Top 10 Pairs, Yahoo! Answers)
Factor 1 Factor 2 Correlation p-value
Honesty First-hand experience 0.604 0.00E+00***
Plausibility Relevance 0.539 0.00E+00***
Plausibility Objectiveness 0.473 0.00E+00*** Objectiveness Comprehensiveness 0.409 0.00E+00***
Objectiveness Relevance 0.365 0.00E+00***
Plausibility Promotion -0.364 0.00E+00***
Comprehensiveness Expertise 0.346 1.78E-15***
Objectiveness Sympathy 0.332 2.58E-14***
Plausibility Negativeness -0.283 1.12E-10***
Positiveness Sympathy 0.269 9.49E-10***
* significant at p < 0.05; ** significant at p < 0.005; *** significant at p < 0.001 More than half of the pairs of credibility factors that were highly correlated to each other were constituted by factors that had a high correlation with credibility (Table 14). For instance, the correlation between plausibility and objectiveness was ranked as third, and both factors were highly correlated with credibility. Objectiveness and comprehensiveness also had a strong correlation with credibility.
However, different patterns were also found in this correlation analysis. In the case of honesty and first-hand experience, which had the highest correlation between factors, both factors had a weak correlation with credibility. This may have happened due to the definitions of
both factors in the coding book (Appendix 8). The definition of honesty is "The answer should be related to the answerer's first-hand experiences and should reveal potential limitations of his/her experience to be considered honest." Since there are limitations for coders to determine whether the author is honest or not, the narrow definition of honesty was used to make the evaluation plausible.
Positiveness and sympathy were also ranked in the top 10 despite their weak correlations with credibility. Having a positive sentiment and having sympathy are different, but there is also an intersection between them. It is rare for a person with sympathy to reveal a negative
sentiment. These factors are not mutually exclusive, but they are different enough to be kept as separate factors. However, there might be potential interactions between these factors. Therefore, interactions between these factors were more carefully examined in later statistical analyses. Yelp
Studying the correlation between credibility and underlying factors with the Yelp dataset also revealed substantial insight. The correlation coefficients and p-values are summarized in Table 15. The credibility factors in Table 15 are listed in descending order according to the correlation coefficient, and the leftmost column explains the social media type to which a corresponding credential factor is applicable. Objectiveness showed the largest positive
correlation (0.49, p < .001) with the credibility, while promotion showed a negative correlation (- 0.01, p < .05). Unlike Yahoo! Answers, no underlying factor that has a strong correlation with credibility was found in Yelp. Four underlying factors, including objectiveness, had a moderate positive relationship with credibility. All other underlying factors showed a weak
positive/negative relationship with credibility. Fewer than half factors (6 out of 14) showed statistically significant differences in Yelp, whereas all factors except for one showed
statistically significant differences in Yahoo! Answers. Regarding the average of absolute values of correlation coefficients in two social media types, Yahoo! Answers were 0.31, more than twice as high as Yelp's 0.14.
Table 15. Correlation between Credibility and Underlying Factors (Yelp) Social Media Type Credibility Factor Correlation p-value
Unique (Yelp)
Specificity 0.41 0.00E+00*** Excessive Positiveness 0.08 8.19E-02 Excessive Negativeness 0.03 5.04E-01
Both (Yahoo! Answers and Yelp) Objectiveness 0.49 0.00E+00*** Honesty 0.35 2.22E-16*** Comprehensiveness 0.31 1.41E-12*** Currency 0.14 1.77E-03** Promotion -0.10 2.54E-02* Credence 0.06 1.70E-01 Presentation 0.05 2.34E-01 Positiveness 0.04 3.45E-01 Negativeness 0.03 5.08E-01 Reference 0.01 7.69E-01 Expertise 0.00 9.68E-01
* significant at p < 0.05; ** significant at p < 0.005; *** significant at p < 0.001 The correlations between pairs of credibility factors were low and not statistically
significant, indicating weak co-occurrence patterns. Correlation and statistical significance of the Yelp dataset were even weaker compared to the Yahoo! Answers dataset. There were two pairs of credibility factors that had moderate correlation, and other pairs had weak correlation between
them. The top 10 correlation pairs were reported in Table 16 for comparison between Yahoo! Answers and Yelp.
Table 16. Correlations between Credibility Factors (Top 10 Pairs, Yelp)
Factor 1 Factor 2 Correlation p-value
Negativeness Excessive
Negativeness -0.428 0.00E+00***
Specificity Honesty 0.301 6.07E-12***
Comprehensiveness Objectiveness 0.295 1.59E-11*** Objectiveness Specificity 0.248 1.94E-08*** Negativeness Objectiveness 0.242 4.22E-08***
Specificity Excessive
Negativeness 0.240 5.53E-08*** Negativeness Excessive Positiveness -0.209 2.48E-06***
Excessive Positiveness
Excessive
Negativeness -0.213 1.61E-06*** Positiveness Excessive Positiveness -0.262 2.68E-09***
Objectiveness Honesty 0.196 9.76E-06***
* significant at p < 0.05; ** significant at p < 0.005; *** significant at p < 0.001
Four pairs of credibility factors that were highly correlated to each other were constituted by factors that had a high correlation with credibility (Table 16). For instance, the correlation between specificity and honesty was ranked as second, and both factors were highly correlated with credibility. Four other pairs of credibility factors had a high correlation due to the
definitions and characteristics of sentiment-related factors. Different degrees of the same sentiment were considered mutually exclusive in the content analysis (Appendix 8). In other words, negativeness cannot occur with excessive negativeness according to the final coding scheme. Also, it is unusual for a very positive review to contain very negative perspectives at the
same time. Therefore, excessive positiveness and excessive negativeness were found to be negatively correlated. Although simple correlation analysis is limited in revealing complex interactions between factors, interactions between these factors can have negative effects on the performance of credibility models. Potential interactions between these factors were carefully examined in later statistical analyses.
4.4.2 Regression Analysis