I am conducting a multilevel logistic regression to identify determinants of the availability of diagnostic tests.
I am using a multilevel approach because I want to examine the effects of medical surgeries’ characteristics (level 1) and the effects of country (level 2) on the availability of the test in the surgery.
I am using 5 surgery explanatory variables (three are binary variables and two are continuous variables) and two country explanatory variables (both categorical). I have 847 observations (i.e. 847 surgeries) and 19 countries. The continuous variables are centered to the mean.
I am using the nlcom command to obtain the country variance after the melogit command.
My problem is that I obtain a negative lower boundary for the 95 %CI of the variance, but I understand the variance cannot be negative.
Does anyone have a suggestion about why could this happen?
Is there another command in stata instead of nlcom that I could use to avoid having negative boundaries for variance?
Thank you !
Code:
melogit dipstick_available_for_use ib2.public_private2 ib2.group_solo2 centered_distance_lab99 centered_pc_turna_time3 i.bloods_taker3 ib3.expend_capita_quart i.source_financing2 [pweight=scaled_pweight] || country:
Code:
Mixed-effects logistic regression Number of obs = 847 Group variable: country Number of groups = 19 Obs per group: min = 3 avg = 44.6 max = 162 Integration method: mvaghermite Integration pts. = 7 Wald chi2(10) = 62.55 Log pseudolikelihood = -223.54672 Prob > chi2 = 0.0000 (Std. Err. adjusted for 19 clusters in country) --------------------------------------------------------------------------------------------------- | Robust dipstick_available_for_use | Coef. Std. Err. z P>|z| [95% Conf. Interval] ----------------------------------+---------------------------------------------------------------- public_private2 | Private | .6230582 1.365527 0.46 0.648 -2.053326 3.299443 | group_solo2 | Solo practice. I am the only h.. | .4136781 .5851945 0.71 0.480 -.7332821 1.560638 | centered_distance_lab99 | -.036124 .031088 -1.16 0.245 -.0970553 .0248072 centered_pc_turna_time3 | .1929808 .1399966 1.38 0.168 -.0814074 .467369 3.bloods_taker3 | -.8310664 .6179098 -1.34 0.179 -2.042147 .3800147 | expend_capita_quart | 0- | -3.333408 1.727199 -1.93 0.054 -6.718656 .0518407 2400- | -1.755793 1.432835 -1.23 0.220 -4.564098 1.052511 2900- | -2.2454 1.490375 -1.51 0.132 -5.16648 .675681 | source_financing2 | gov | 1.308304 .7677665 1.70 0.088 -.196491 2.813098 vhi_oop | 1.075064 1.279517 0.84 0.401 -1.432742 3.582871 | _cons | 4.341836 1.405842 3.09 0.002 1.586436 7.097237 ----------------------------------+---------------------------------------------------------------- country | var(_cons)| 1.22608 1.076046 .2195218 6.847939 ---------------------------------------------------------------------------------------------------
Code:
nlcom _b[/var(_cons[country])], cformat(%9.3f)
Code:
_nl_1: _b[/var(_cons[country])] ------------------------------------------------------------------------------ dipsti~r_use | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- _nl_1 | 1.226 1.076 1.14 0.255 -0.883 3.335 ------------------------------------------------------------------------------
0 Response to Negative lower confidence interval boundary for variance with nlcom after a multilevel logistic regression
Post a Comment