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

Table 2 Summary of regression model performance across all populations

	Model	Weight	Clusters	Ψ	SE Adj.	Error	Coverage	Bias (mean %)	Bias (median %)	Accuracy (%)
Logistic Regression
Generalised Linear Models
glm(R)	1	–				0.04	0.954	2.07	−1.63	88.1
	2	RDS-II				0.55	0.442	20.89	8.51
	3	–	R-y			0.04	0.955	3.35	−0.48	88.6
	4	RDS-II	R-y			0.55	0.443	25.56	11.57
surveylogistic (SAS)	5	–				0.05	0.952	2.07	−1.63	88.1
	6	RDS-II				0.07	0.903	20.88	8.51
	7	–			Morel	0.05	0.953	2.07	−1.63	88.1
	8	RDS-II			Morel	0.07	0.904	20.88	8.51
	9	RDS-II	RwS			0.07	0.903	20.88	8.51
	10	RDS-II	RwS		Morel	0.07	0.904	20.88	8.51
Generalised Linear Mixed Models
glmer(R)	11	–	S	U		0.05	0.954	3.48	−0.46	88.1
glmer(R)	12	RDS-II	S	U		0.55	0.402	44.55	26.73
glimmix (SAS)	13	–	S	AR		0.04	0.955	3.45	−0.34	88.1
glimmix (SAS)	14	–	R	CS		0.04	0.957	2.4	−1.19	88.1
glmmPQL(R)	15	–	S	DC		0.04	0.865	−0.86	−6.34
Generalised Estimating Equations
geeglm(R)	16	–	R	I	Classical	0.13	0.952	2.07	−1.63
geeglm(R)	17	RDS-II	R	I	Classical	0.16	0.902	20.89	8.51
glimmix (SAS)	18	–	S	AR		0.04	0.939	1.85	−1.69
	19	–	R	CS		0.04	0.937	2.52	−1.75
	20	–	R	CS	Classical	0.05	0.948	2.52	−1.75
	21	–	R	CS	FIRORES	0.05	0.950	2.52	−1.75	88.1
	22	–	R	CS	FIROEEQ	0.05	0.951	2.52	−1.75	88.1
	23	–	R	CS	MBN	0.05	0.950	2.52	−1.75
Poisson Regression
Generalised Linear Models
glm(R)	24	–				0.02	0.962	4.81	4.15	86
glm(R)	25	RDS-II				0.49	0.457	9.48	8.23
glm(R)	26	–	R-y			0.02	0.964	3.06	2.44	86.3
glm(R)	27	RDS-II	R-y			0.47	0.493	7.74	6.46
Generalised Linear Mixed Models
glmer(R)	28	–	S	U		0.02	0.963	4.92	4.27	86
glmer(R)	29	RDS-II	S	U		0.47	0.431	11.71	10.42
Generalised Estimating Equations
geeglm(R)	30	–	R	I	Classical	0.13	0.859	4.81	4.15
geeglm(R)	31	RDS-II	R	I	Classical	0.17	0.781	9.48	8.23

R-y recruiter outcome as covariate, S Seeds, R recruiter, RwS recruiter within seed