Modelling count, bounded and skewed continuous outcomes in physical activity research: beyond linear regression models

Table 2 Estimated associations between PA and NCMC adjusted for age and sex from the ALECS study of physical activity in Hong Kong older adults using various regression models (N = 402)

	LM	GLM-Poisson	GLM-quasi-Poisson	GLM-Negative Binomial
Intercept
Estimate	875.5	6.918	6.918	7.021
SE	349.1	0.027	0.645	0.624
p-value	0.013	< 0.001	< 0.001	< 0.001
95% CI	189.2, 1562	6.87, 6.97	5.66, 8.19	5.77, 8.25
Age (years)
Estimate	-3.997	-0.007	-0.007	-0.009
SE	4.668	0.0004	0.009	0.008
p-value	0.392	< 0.001	0.388	0.305
95% CI	-13.17, 5.18	-0.008, -0.007	-0.025, 0.009	-0.025, 0.008
Sex (Female reference)
Estimate	82.320	0.151	0.151	0.149
SE	61.365	0.005	0.111	0.110
p-value	0.181	< 0.001	0.173	0.175
95% CI	-38.32, 203.0	0.14, 0.16	-0.07, 0.37	-0.06, 0.37
NCMC
Estimate	-20.078	-0.038	-0.038	-0.046
SE	14.180	0.001	0.027	0.025
p-value	0.158	< 0.001	0.153	0.072
95% CI	-47.95, 7.80	-0.041, -0.036	-0.091, 0.014	-0.099, 0.009
AIC	6244.3	183,071	5873.9	5863.4
DP	-	-	586.3	0.976

SE: standard error; CI: confidence interval; LM: linear regression model; NCMC: number of chronic medical conditions; AIC: Akaike information criterion; DP: dispersion parameter. All analyses were conducted in R version 4.0.2 [45]

ISSN: 1479-5868