Stepwise regression within multiply imputed data - problem with loop

[Josefie Breedvelt]

I am conducting stepwise regression within multiply imputed data, using the approach as specified by Wood, White & Royston (2008) https://onlinelibrary.wiley.com/doi/....1002/sim.3177

I have ran my multiple imputation successfully, creating 20 datasets. Please see code below.

The challenge arises when I attempt to use the Wald method for variable selection within my dataset. Here, I fit the model with all predictors in each dataset, then I calculate the wald statistic for each predictor and pool these using Rubin's Rules. Then for each dataset I want to identify the variable with the smallest wald chi2, get it's p value and remove it from the model if it exceeds p = 0.1. I then want to repeat the process in all imputed datasets until all remaining variables meet my p value criterion.

When I create my loop to evaluate the wald statistics, two errors arise:

Errors:
1, gen dataset = `m'
invalid syntax
r(198);
2. . }
} is not a valid command name
r(199);

What am I doing wrong?

Loop code:
forval `m' = 1/20 {
preserve
di "Running Poisson Regression for Imputed dataset `m'"
mi extract`m'
poisson diagnos female ethn anx medinc bully sociso


matrix b_vec = e(b)

mat v = e(V)

mat list v

mata:

b = st_matrix("b_vec")

V = st_matrix("v")

V = diagonal(V)

se_t = sqrt(V)

se = se_t'

b

se

wald = b:/se

st_matrix("wald", wald)

end

matrix list wald

svmat wald

keep wald*

drop if wald1 == .

gen dataset = `m'

save wald_results_`m'.dta, replace
}



Multiple imputation:
mi set wide
mi register imputed diagnos female ethn anx medinc bully sociso
mi impute chained (logit) diagnos female ethn anx medinc bully sociso, add(20) rseed(12345)

[Andrew Musau]

forval `m' = 1/20 {

"m" here is an index and not a macro, so you need:

forval m=1/20{

[Josefie Breedvelt]

"m" here is an index and not a macro, so you need:

forval m=1/20{

Thank you Andrew! I have re-ran the code and it still gives an error for gen dataset = m or gen dataset = `m' - so the output I get is just one dataset that is being created with the wald results of one study.

gen dataset=m
m not found
r(111);

Anything I am doing wrong? Thank you

[Andrew Musau]

You need the macro quotes in this instance. It appears you are saving 20 datasets, so search for these.

preserve
forval m=1/20{
    restore, preserve
    di "Running Poisson Regression for Imputed dataset `m'"
    mi extract`m'
    poisson diagnos female ethn anx medinc bully sociso

    matrix b_vec = e(b)
    mat v = e(V)
    mat list v

    mata:
    b = st_matrix("b_vec")
    V = st_matrix("v")
    V = diagonal(V)
    se_t = sqrt(V)
    se = se_t'
    b
    se
    wald = b:/se
    st_matrix("wald", wald)
    end

    matrix list wald

    svmat wald
    keep wald*
    drop if wald1 == .
    gen dataset = `m'
    save wald_results_`m'.dta, replace
}