Closed jay-sf closed 1 year ago
Currently, when we impute datasets, the results are different depending on the column order.
> library(mice) > summary(pool(with(mice(nhanes2, seed=42, printFlag=FALSE), lm(chl ~ age + hyp + bmi)))) term estimate std.error statistic df p.value 1 (Intercept) -7.041069 68.984151 -0.1020679 10.048167 0.92071091 2 age40-59 44.738210 22.610459 1.9786511 12.654316 0.07004070 3 age60-99 71.304537 27.001228 2.6407887 9.457629 0.02581795 4 hypyes -5.289385 22.824540 -0.2317411 10.452068 0.82121333 5 bmi 6.511697 2.560452 2.5431830 8.172416 0.03396548 > summary(pool(with(mice(nhanes2[, c(2, 4, 1, 3)], seed=42, printFlag=FALSE), lm(chl ~ age + hyp + bmi)))) term estimate std.error statistic df p.value 1 (Intercept) 41.897815 65.434937 0.64029732 11.175928 0.53488653 2 age40-59 46.541989 22.261161 2.09072602 13.942022 0.05535303 3 age60-99 62.623802 25.442422 2.46139313 11.357257 0.03097931 4 hypyes -3.134964 32.152478 -0.09750303 4.631723 0.92640397 5 bmi 4.621302 2.440856 1.89331180 9.160326 0.09028175
It is suspected that the internal use of seeds might be the reason for this. Also see related Q&A on Stack Overflow.
See https://stackoverflow.com/a/75694675/21369628 for an explanation. This is expected behavior, so the issue can be closed.
Currently, when we impute datasets, the results are different depending on the column order.
It is suspected that the internal use of seeds might be the reason for this. Also see related Q&A on Stack Overflow.