Running Head: Invisible Leverage in Adoption of Online Social Support Community

20

In general, simultaneous third-order polynomial regression analyses of all three network

size measures found that the contribution of the number of adopters to R-square change was the

highest, the number of posters the next highest, and the number of registered members was the

lowest.

Hypothesis 3: The network threshold of physical illness OSSCs tends to be larger than that of

mental illness OSSCs.

Hypothesis 3 was not supported. When separate third-order polynomial regressions were

performed by illness type, no significant curvilinear relationship was found in any measure of

adoption on both illness types at the same time. Therefore, no comparison of network thresholds

by illness type was possible.

Hypothesis 4: There will a curvilinear relationship between the participation of the major

contributor and adoption.

Hypothesis 4-1 was partially supported. The major contributor’s frequency activity

showed curvilinear as well as positive linear relationships with four adoption measures – the

frequency of postings per adopter, the frequency of postings per adopter excluding the major

contributor, the proportion of adopters among posters, and network activeness. Total variances

in all these four measures were better explained by curvilinear regression than by linear

regression (See Table 4).

The major contributor’s frequency activity accounted for 30.6% of the total variance

(p<.01) in terms of the frequency of postings per adopter, for 18.0% (p<.01) in terms of the

frequency of postings per adopter excluding the major contributor, for 8.1% (p<.01) in terms of

the proportion of adopters among posters, and for 35.1% (p<.01) in terms of network activeness

(all curvilinear).

Hypothesis 4-2 was also partially supported. The major contributor’s duration activity

showed curvilinear as well as positive linear relationships with the average duration and the

average duration excluding the major contributor. Total variances in the two measures were

better explained by curvilinear regression than by linear regression. Only a curvilinear

relationship was found with the proportion of adopters among posters. However, unlike the

findings of the previous hypotheses, standardized coefficients for quadratic terms were negative,