Skip to main content
. 2023 Jul 18;16(5):555–567. doi: 10.1007/s40271-023-00640-z

Table 3.

Conditional logit and mixed logit estimates

Attributes and levels Conditional logit Mixed logit
Mean SD
Coeff SE Coeff SE Coeff SE
Contribution to antibiotic resistance
Decrease by a quarter 0.411 [0.028]** 0.710 [0.054]** 0.026 [0.064]
Stay at the current level (Ref.) 0.189 [0.024]** 0.300 [0.042]**
Double −0.162 [0.025]** −0.244 [0.040]** 0.133 [0.125]
Triple −0.438 [0.030]** −0.766 [0.058]** 0.680 [0.061]**
Treatment duration
3 Days 0.075 [0.021]** 0.132 [0.035]** 0.504 [0.050]**
7 Days 0.021 [0.018] 0.050 [0.030] 0.024 [0.041]
14 Days (Ref.) −0.096 [0.021]** −0.182 [0.035]**
Side effects
1% 0.175 [0.019]** 0.287 [0.033]** 0.505 [0.049]**
10% −0.022 [0.019] −0.007 [0.030] 0.147 [0.142]
25% (Ref.) −0.153 [0.022]** −0.280 [0.038]**
Days needed to recover
3–7 Days 0.146 [0.019]** 0.246 [0.034]** 0.353 [0.071]**
8–14 Days −0.025 [0.019] −0.026 [0.031] 0.361 [0.063]**
15–28 Days (Ref.) −0.121 [0.021]** −0.220 [0.036]**
Days before you start your medication
Straight after seeing a GP 0.048 [0.013]** 0.082 [0.022]** 0.305 [0.043]**
Wait 3 days (Ref.) −0.048 [0.013]** −0.082 [0.022]**
Treatment failure
5% 0.045 [0.018]* 0.098 [0.031]** 0.343 [0.062]**
10% 0.086 [0.017]** 0.128 [0.028]** 0.212 [0.063]**
25% (Ref.) −0.130 [0.020]** −0.225 [0.034]**
Out-of-pocket costs
$10 0.396 [0.025]** 0.779 [0.058]** 1.096 [0.067]**
$30 0.109 [0.017]** 0.175 [0.028]** 0.002 [0.033]
$70 (Ref.) −0.505 [0.026]** −0.954 [0.063]**
Observations 29,844 29,844
N 1658 1658
LL −9006 −8375
BIC 18157 17038

Notes: **p < 0.01, *p < 0.05. Robust standard errors (SEs) are reported in the table. Effect coding was used for all attributes. For the mixed logit model, all attribute levels were assumed to be random variables. SD, standard deviation. The number of Halton draws used for the simulation: 1000. LL, log likelihood; BIC, Bayesian information criterion. A post-stratification weight (calculated according to age groups and gender distribution) was applied

There was inconsistency between estimated coefficients of 5% and 10% but they are statistically indifferent from each other