“3.4.3.” bölümünde [?] kısmımlarına ve Table E.3.’deki [?] ...

Prompt

“3.4.3.” bölümünde [?] kısmımlarına ve Table E.3.’deki [?] kısımlarına ne gelir? 3.4.3. Sensory Cue Count and Engagement Among the 84 videos with complete metrics, SII was positively associated with follower-normalised engagement (Spearman's ρ = 0.58, 95% BCa CI [0.43, 0.70], q < 0.001). The association remained positive in the adjusted model (b=0.118b = 0.118b=0.118, 95% CI [0.050, 0.186], β=0.391\beta = 0.391β=0.391, q=0.001) and in outlier-excluded, multiple-imputation, and negative-binomial specifications (IRR = 1.11, 95% CI [1.04, 1.19]; Appendix E, Table E.2). In the full pooled interaction model, the estimated SII slope was b = 0.118 for Instagram (the reference category) and b = 0.145 for TikTok (0.118 + 0.027). The Platform × SII interaction was positive but not detectable (b = 0.027, 95% CI [−0.110, 0.164], p = 0.700); accordingly, the data provide no evidence that the SII–engagement slope differs between platforms. Separate platform-specific models with a reduced covariate set yielded b = 0.124 for Instagram (n = 61, 95% CI [0.041, 0.207]) and b = 0.097 for TikTok (n = 23, 95% CI [−0.012, 0.206]); these estimates are reported as sensitivity analyses rather than as tests of effect modification. The TikTok interval reflects greater uncertainty, not absence of association. Because publication dates were available for most videos, video age was computable for approximately 81–82 videos. A sensitivity model additionally adjusted for ln(video age); among the 81–82 videos with complete engagement and publication-date data, the SII coefficient was b = [?] (95% CI [?], p = [?]). This specification addresses the possibility that older videos accumulated more engagement, although it does not model exposure directly. Because the sample was restricted to high-engagement videos (Section 2.2) and exposure measures such as reach, watch time, and video age were unavailable, the association describes co-variation among already successful content and cannot establish that adding sensory cues increases engagement. ++++ Table E.3. Sensitivity of the Health/Pleasure Comparison Scenario n pleasure / health M pleasure (SD) M health (SD) Welch t (df) p Hedges' g [95% BCa CI] S1 Main analysis 61 / 28 0.71 (0.29) 0.29 (0.36) 5.58 (45.17) < 0.001 1.32 [0.86, 1.77] S2 Hashtag criterion removed from classification 59 / 30 0.70 (0.30) 0.30 (0.33) 5.41 (47.30) < 0.001 1.27 [0.80, 1.73] S3 34 health/pleasure words removed from lexicon 61 / 28 0.62 (0.29) 0.28 (0.32) 4.57 (44.92) < 0.001 1.08 [0.62, 1.53] S4 Unclassifiable videos added to health group 61 / 44 0.71 (0.29) 0.40 (0.33) 4.81 (88.61) < 0.001 0.98 [0.56, 1.39] S5 English-only comment corpora 38 / 17 0.73 (0.29) 0.27 (0.35) 4.52 (25.84) < 0.001 1.36 [0.76, 1.94] S6 Turkish-only comment corpora 19 / 9 0.68 (0.30) 0.31 (0.34) 2.92 (14.66) 0.007 1.18 [0.38, 1.95] Primary model + ln⁡(video age)ln(video age) [81/82] b=[?] [?] [?] [?] Primary model + ln⁡(video age)ln(video age) Note. M = video-level mean comment sentiment (range −1 to +1). In S3, words common in health and pleasure contexts (e.g., "fresh," "fatty," "sugary," "light") were removed in both pipelines; this directly affects lexicon-based VADER but only partly affects contextual BERTurk. S4 is the most conservative scenario. S5–S6 have smaller groups because videos with bilingual comment corpora were excluded. Only S1 was included in family A5. Video age was calculated from the verified publication date and entered as ln⁡(video age)ln(video age). This sensitivity model was restricted to videos with complete engagement metrics and a verifiable publication date; it adjusts for elapsed time since publication but does not measure cumulative exposure, reach, or watch time.