Hello,
I am new to multiple imputation and have some doubt regarding the imputed models
I am currently working on a lifecourse epidemiology dataset (n=1520) wherein the variables are 0-70% missing.
I ran the below mentioned
mi set mlong
mi register imputed Glucose HDL Triglyceride wc alcohol_gms mother_age mean_SBP_DBP wt0 wt1 wt2 wt3 wt4 wt5 wt6 g_res* breastfed BMI PA MET_Syndrom mother_bmi smoker ses ht0 ht1 ht2 ht3 ht4 ht5 ht6 region
mi impute chained (pmm, knn(10)) ses wc alcohol_gms mother_age Triglyceride BMI PA mother_bmi (regress) mean_SBP_DBP wt0 wt1 wt2 wt3 wt4 wt5 wt6 HDL Glucose GA g_res* wt0 ht0 ht1 ht2 ht3 ht4 ht5 ht6 (logit) smoker breastfed MET_Syndrom = age school sex study, add(250) rseed (53421) savetrace (trace1, replace)
mi estimate: regress BMI g_res* alcohol_gms mother_age GA i.breastfed PA ses mother_bmi i.smoker age i.sex
My endpoints are BMI Triglyceride Glucose HDL, though the above model runs perfectly, but I also wanted to see the association with dichotmous variables of BMI and others, which I couldnt add due to perfect prediction error. Instead of adding the augment option I considered getting rid of some of the variables. My question is can I have more than 1 imputed model of the same dataset each time with different dichotmous variable. Theoretically, it is possible but will it be correct to do it or is there any other way to improve my model.
Further when I checked my highest RVI was 5.26 for mother_bmi. I have no other variable which I could add in the imputaion model which is correlated with mother_bmi to decrease the variance.
It would be great to have suggestions on this
I am new to multiple imputation and have some doubt regarding the imputed models
I am currently working on a lifecourse epidemiology dataset (n=1520) wherein the variables are 0-70% missing.
I ran the below mentioned
mi set mlong
mi register imputed Glucose HDL Triglyceride wc alcohol_gms mother_age mean_SBP_DBP wt0 wt1 wt2 wt3 wt4 wt5 wt6 g_res* breastfed BMI PA MET_Syndrom mother_bmi smoker ses ht0 ht1 ht2 ht3 ht4 ht5 ht6 region
mi impute chained (pmm, knn(10)) ses wc alcohol_gms mother_age Triglyceride BMI PA mother_bmi (regress) mean_SBP_DBP wt0 wt1 wt2 wt3 wt4 wt5 wt6 HDL Glucose GA g_res* wt0 ht0 ht1 ht2 ht3 ht4 ht5 ht6 (logit) smoker breastfed MET_Syndrom = age school sex study, add(250) rseed (53421) savetrace (trace1, replace)
mi estimate: regress BMI g_res* alcohol_gms mother_age GA i.breastfed PA ses mother_bmi i.smoker age i.sex
My endpoints are BMI Triglyceride Glucose HDL, though the above model runs perfectly, but I also wanted to see the association with dichotmous variables of BMI and others, which I couldnt add due to perfect prediction error. Instead of adding the augment option I considered getting rid of some of the variables. My question is can I have more than 1 imputed model of the same dataset each time with different dichotmous variable. Theoretically, it is possible but will it be correct to do it or is there any other way to improve my model.
Further when I checked my highest RVI was 5.26 for mother_bmi. I have no other variable which I could add in the imputaion model which is correlated with mother_bmi to decrease the variance.
It would be great to have suggestions on this
