Unweighted regression models perform better than weighted regression techniques for respondent-driven sampling data: results from a simulation study

Table 4 Type I error rate of unweighted and weighted regression models for populations with correlation between outcome and network degree

Secondary analysis population		Binomial regression		Poisson regression
Correlation of degree and outcome		Unweighted	Weighted	Unweighted	Weighted
Population homophily = 1.25
1	extreme negative (ρ = −0.133)	0.043	0.548	0.037	0.455
2	extreme positive (ρ = 0.534)	0.048	0.003	0.037	0.003
3	moderate negative (ρ = −0.092)	0.062	0.498	0.049	0.445
4	moderate positive (ρ = 0.059)	0.046	0.241	0.032	0.229
Population homophily = 1.50
5	extreme negative (ρ = −0.132)	0.037	0.529	0.029	0.412
6	extreme positive (ρ = 0.534)	0.054	0.006	0.043	0.006
7	moderate negative (ρ = −0.093)	0.037	0.459	0.025	0.418
8	moderate positive (ρ = 0.060)	0.024	0.186	0.020	0.175

ISSN: 1471-2288