Stated preferences for anti-malarial drug characteristics in Zomba, a malaria endemic area of Malawi

Malaria Journal

Table 2 Discrete choice model estimates by model variant (Linear index - Equation 2)

	Model 1: Normally distributed coefficients (N=508 respondents, 3899 responses)		Model 2: Log-normally distributed coefficients (N=508 respondents, 3899 responses)
Variable	Estimate Coefficient [SD]	Standard error Coefficient [SD]	Estimate Coefficient [SD]	Standard error Coefficient [SD]
Cost($)	−0.63 [0.52]	0.05* [0.06]*	n/a	n/a
Fever duration¹	−0.11 [0.03]	0.01*** [0.03]	n/a	n/a
Risk of rash	−1.27 [0.76]	0.09* [0.10]*	n/a	n/a
Savings($) (i.e. inverse of cost)	n/a	n/a	−0.76 [1.59]	0.14* [0.38]*
Days without fever (i.e. inverse of fever duration)	n/a	n/a	0.015 [0.02]	0.011 [0.02]
Probability of no side effect (i.e. inverse of risk of rash)	n/a	n/a	0.14 [0.90]	0.09 [0.13]***
Official treatment	3.06 [2.29]	0.24* [0.22]*	3.44 [2.42]	0.32* [0.32]*
Prophylaxis duration	0.05 [0.04]	0.004* [0.005]*	0.06 [0.04]	0.005* [0.006]*
Course duration	−0.29 [0.06]	0.05*** [0.11]	−0.32 [0.20]	0.06*** [0.13]
Relative variance: Urban vs. rural¹	1.59 [0]	0.37	1.43 [0]	0.30
Relative variance: Peri-urban vs. rural¹	0.91 [0]	0.19	0.85 [0]	0.19
−2 x(Mean simulated Log-likelihood ratio)²	47.4***		10.40

Note: only coefficient estimates with asterisks alongside their standard errors had p<0.05.
*** p<0.001. n/a: not included in model.
1. Relevant test for relative variances is Ho: Variance urban/ Variance rural =1, and Variance peri-urban/ Variance rural=1, so p>0.05 (for urban vs. rural and peri-urban vs. rural, separately, and, or jointly using the simulated log-likelihood ratio tests –results not presented but available from authors).
2. Quasi-likelihood ratio test of the null hypothesis that all attribute fixed coefficients being equal to 0 (Note that in the case of Model 2, where three coefficients are log normally distributed –for cost, fever resolution, and risk of rash- this hypothesis is equivalent to their being jointly insignificantly different from 1, along with the normally distributed coefficients being zero). The test statistics are distributed in both cases as χ(6 degrees of freedom).
Model was estimated using simulation with n=100 repetitions.

ISSN: 1475-2875