Table 5

Regression results: predictors of negative sentiment and engagement (HC1 robust SEs)

PredictorModel 1: Negativity b (SE) [95% CI]Model 2: Engagement b (SE) [95% CI]p (M1)p (M2)
Delay & Disruption0.036 (0.001)
[0.034, 0.038]
−0.909 (0.133)
[−1.170, −0.648]
<0.001<0.001
Crowding & Capacity−0.001 (0.002)
[−0.005, 0.003]
0.311 (0.213)
[−0.107, 0.729]
0.4340.144
Ticketing & Refunds−0.010 (0.001)
[−0.012, −0.008]
−0.956 (0.126)
[−1.203, −0.709]
<0.001<0.001
Comfort & Cleanliness0.002 (0.001)
[0.000, 0.004]
0.635 (0.122)
[0.396, 0.874]
0.007<0.001
Staff & Communication0.005 (0.001)
[0.003, 0.007]
0.250 (0.151)
[−0.046, 0.546]
<0.0010.098
Safety & Security0.044 (0.002)
[0.040, 0.048]
0.818 (0.233)
[0.361, 1.275]
<0.001<0.001
Accessibility−0.002 (0.002)
[−0.006, 0.002]
0.075 (0.318)
[−0.548, 0.698]
0.2460.815
Delay duration (log min)−0.000 (0.001)
[−0.002, 0.002]
−0.162 (0.063)
[−0.285, −0.039]
0.8300.010
Controls (record type, subreddit, length, year)includedincluded
R2/N0.028/61,3280.325/2,173

Note(s): Intervals are Wald 95% confidence intervals computed as b ± 1.96 SE from the coefficients and standard errors reported above, both of which are rounded to three decimals; the limits are therefore accurate only to the precision of those inputs and narrow intervals may appear marginally inconsistent with the reported p-values. Because n = 61,328, conventional significance is attained by effects of negligible magnitude: Model 1 accounts for 2.8% of the variance in document-level negativity, so the substantive reading of this table rests on the relative ordering and magnitude of the coefficients rather than on their significance levels and no coefficient should be interpreted as establishing a practically important difference on the strength of its p-value alone.

Source(s): Authors’ own work

or Create an Account

Close subscription notice
Close access options