Hello,
I am analyzing individual level data from a cluster RCT. There are 24 clusters (12 intervention & 12 control clinics) and the number of patients recruited from each clinic is between 84 and 150. The total number of patients is 2796. In this secondary analysis I am not comparing between the study arms but pooling the clusters to look at factors associated with persistent clinical depression. My main outcome is a binary variable (persistent depression during one year follow up: Yes/No) Since the number of clusters are fewer than ideal, I am using the score bootstrap developed by Kline and Santos as an adaptation of the wild bootstrap. (Kline, P., and Santos, A. 2012. A score based approach to wild bootstrap inference. Journal of Econometric Methods 1(1): 23-41.).
To implement this, I am using the boottest (ssc boottest) package developed by Roodman et al. (Roodman, D., J. MacKinnon, M. Nielsen, and M. Webb. 2018. Fast and wild: bootstrap inference in Stata using boottest. Queen's Economics Department Working Paper No. 1406 ) in Stata IC/14.2
While running a multivariable model I am getting an error:
Here is the command that I used
after running the logistic regression and rerunning with the score bootstrap, null imposed, it runs fine for the first two independent variables. Here is the output for the logistic regression and the independent variable sex_b
But once it reaches the third variable I am getting the following error:
I would really appreciate any thoughts on what is causing this error.
OR any other suggestions for analysis of binary outcome variable in a data with few clusters (24 clusters & The ICC for my outcome variable is about 0.082)
Thank you
I am analyzing individual level data from a cluster RCT. There are 24 clusters (12 intervention & 12 control clinics) and the number of patients recruited from each clinic is between 84 and 150. The total number of patients is 2796. In this secondary analysis I am not comparing between the study arms but pooling the clusters to look at factors associated with persistent clinical depression. My main outcome is a binary variable (persistent depression during one year follow up: Yes/No) Since the number of clusters are fewer than ideal, I am using the score bootstrap developed by Kline and Santos as an adaptation of the wild bootstrap. (Kline, P., and Santos, A. 2012. A score based approach to wild bootstrap inference. Journal of Econometric Methods 1(1): 23-41.).
To implement this, I am using the boottest (ssc boottest) package developed by Roodman et al. (Roodman, D., J. MacKinnon, M. Nielsen, and M. Webb. 2018. Fast and wild: bootstrap inference in Stata using boottest. Queen's Economics Department Working Paper No. 1406 ) in Stata IC/14.2
While running a multivariable model I am getting an error:
Here is the command that I used
Code:
logit cisr_hi2 sex_b age_b edu_r1 _Iman_fina__3 ill_r1 marit3 , cluster (clinic_b) or //
boottest {sex_b = 0} { age_b = 0} { edu_r1 = 0} {_Iman_fina__3 = 0 }{ill_r1 = 0} {marit3 = 0}
after running the logistic regression and rerunning with the score bootstrap, null imposed, it runs fine for the first two independent variables. Here is the output for the logistic regression and the independent variable sex_b
Code:
logit cisr_hi2 sex_b age_b edu_r1 _Iman_fina__3 ill_r1 marit3 , cluster (clinic_b) or
Iteration 0: log pseudolikelihood = -918.72324
Iteration 1: log pseudolikelihood = -874.81109
Iteration 2: log pseudolikelihood = -873.0685
Iteration 3: log pseudolikelihood = -873.06434
Iteration 4: log pseudolikelihood = -873.06434
Logistic regression Number of obs = 2,122
Wald chi2(6) = 120.78
Prob > chi2 = 0.0000
Log pseudolikelihood = -873.06434 Pseudo R2 = 0.0497
(Std. Err. adjusted for 24 clusters in clinic_b)
-------------------------------------------------------------------------------
| Robust
cisr_hi2 | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
sex_b | .6541794 .1518481 -1.83 0.068 .4150649 1.031045
age_b | 1.005826 .0083267 0.70 0.483 .9896371 1.022279
edu_r1 | .7372227 .108622 -2.07 0.039 .5523103 .9840435
_Iman_fina__3 | 2.46528 .3796867 5.86 0.000 1.822928 3.333979
ill_r1 | 1.630592 .2851891 2.80 0.005 1.157366 2.29731
marit3 | .8978088 .1216704 -0.80 0.426 .6883822 1.170949
_cons | .0806981 .0340102 -5.97 0.000 .0353285 .1843325
-------------------------------------------------------------------------------
. boottest {sex_b = 0} { age_b = 0} { edu_r1 = 0} {_Iman_fina__3 = 0 }{ill_r1 = 0} {marit3 = 0}
Re-running regression with null imposed.
Iteration 0: log likelihood = -918.72324
Iteration 1: log likelihood = -876.92721
Iteration 2: log likelihood = -875.38767
Iteration 3: log likelihood = -875.38502
Iteration 4: log likelihood = -875.38502
Logistic regression Number of obs = 2,122
Wald chi2(5) = 80.70
Log likelihood = -875.38502 Prob > chi2 = 0.0000
( 1) [cisr_hi2]sex_b = 0
-------------------------------------------------------------------------------
cisr_hi2 | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
sex_b | 1 (omitted)
age_b | 1.003409 .0057257 0.60 0.551 .9922498 1.014694
edu_r1 | .7059302 .1047639 -2.35 0.019 .5277628 .944245
_Iman_fina__3 | 2.463688 .3145025 7.06 0.000 1.918339 3.164069
ill_r1 | 1.644292 .2119911 3.86 0.000 1.277137 2.116999
marit3 | .9717333 .1368774 -0.20 0.839 .7373058 1.280697
_cons | .0844367 .0253467 -8.23 0.000 .0468827 .1520724
-------------------------------------------------------------------------------
Score bootstrap, null imposed, 999 replications, Wald test, clustering by cluster(clinic_b), bootstrap clustering by clinic_b, Radema
> cher weights:
sex_b = 0
z = -2.4064
Prob>|z| = 0.0320
But once it reaches the third variable I am getting the following error:
Code:
Re-running regression with null imposed.
Iteration 0: log likelihood = -918.72324
Iteration 1: log likelihood = -876.75676
Iteration 2: log likelihood = -875.17568
Iteration 3: log likelihood = -875.17242
Iteration 4: log likelihood = -875.17242
Logistic regression Number of obs = 2,122
Wald chi2(5) = 80.88
Log likelihood = -875.17242 Prob > chi2 = 0.0000
( 1) [cisr_hi2]edu_r1 = 0
-------------------------------------------------------------------------------
cisr_hi2 | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
sex_b | .6201915 .1251833 -2.37 0.018 .4175553 .9211655
age_b | 1.010337 .0054863 1.89 0.058 .9996412 1.021147
edu_r1 | 1 (omitted)
_Iman_fina__3 | 2.558244 .3235625 7.43 0.000 1.996564 3.277936
ill_r1 | 1.591771 .2046632 3.62 0.000 1.237191 2.047975
marit3 | .9076537 .1320017 -0.67 0.505 .6825405 1.207013
_cons | .0588529 .0154306 -10.80 0.000 .035204 .0983885
-------------------------------------------------------------------------------
Score bootstrap, null imposed, 999 replications, Wald test, clustering by cluster(clinic_b), bootstrap clustering by clinic_b, Radema
> cher weights:
edu_r1 = 0
z = -1.9053
Prob>|z| = 0.0791
Re-running regression with null imposed.
_cns_eigen(): 3301 subscript invalid
opt__validate(): - function returned error
mopt__validate(): - function returned error
_moptimize(): - function returned error
Mopt_maxmin(): - function returned error
<istmt>: - function returned error
Error imposing null. Perhaps logit does not accept the constraints(), from(), and iterate() options, as needed.
r(3301);
I would really appreciate any thoughts on what is causing this error.
OR any other suggestions for analysis of binary outcome variable in a data with few clusters (24 clusters & The ICC for my outcome variable is about 0.082)
Thank you
