
Conditional logit model^{a}

Latent class logit model^{b}



Class 1 (High WTA)

Class 2 (Low WTA)


Alternativespecific attributes

WTA

Std. Err.

WTA

Std. Err.

WTA

Std. Err.


Foregone 10 % malaria risk reduction

$8.94***

($3.40)

$19.35***

($5.35)

$0.38

($0.31)

One IRS round foregone

DDTbased

$56.38***

($14.57)

$87.53***

($20.43)

$1.97***

($0.94)

ICONbased

$53.78***

($13.69)

$84.32***

($20.01)

$2.79***

($1.07)

Predicted class sizes

Unconditional
 
80 %

20 %

(w/ sample weights)
 
82 %

18 %

Conditional
 
81 %

19 %

(w/ sample weights)
 
84 %

16 %

Respondents

588

588

Choice tasks per respondent

3

3

Model degrees of freedom

4

25

Loglikelihood

−1 376

−1 166


Notes: ***, ** and * indicate statistical significance at the 1, 5 and 10 % levels, respectively. Computations based on conditional and latent class logit model estimates (Additional file 2: Table A1). A 10 % discount rate is applied to convert the choice model coefficients to annual WTA, according to estimates reported by Bauer and Chytilová [36]. Dollar values in 2009 USD. Standard errors calculated clustering at the respondent level (i.e. across choice tasks). ^{a} Model estimated with sampling weights. Model estimated without sampling weights yields similar results, but with a 34 % lower (in magnitude) malaria risk WTA and a loglikelihood value of −1419. ^{b} Model and reported loglikelihood first estimated without sampling weights, due to software limitations. To account for sampling design, sampling weights applied to class membership model and imputed class sizes reported here with and without sample weights