Metan: Change p value for test for heterogeneity from p=0.000 to p<0.001

[Mel de Lange]

Hi there,

I am running a series of meta analyses in a loop to produce forest plots. Code:

levels VariableName
foreach i in `r(levels)'{
metan beta_ se_ if VariableName =="`i'", lcols(month)
graph save "file_`i'", replace
}

However, the p value for the test of heterogeneity appears as p=0.000 and I've been asked for that to be changed to p<0.001.

I have read a previous thread about this (https://www.statalist.org/forums/for...in-forest-plot) and attempted to adapt the code suggested there but Stata produced the error "variable _EFFECT not found" so I've definitely done something wrong!

How would I get p=0.000 to display as p<0.001 in this instance?

I am using Stata 17.

Many thanks in advance,

Mel

[George Ford]

the display is in the "metan_output.ado" at line 731.

try changing to this.

                        if `pvalue'>0.001 {
                            disp as text "  p = " as res %5.3f `pvalue'
                        }
                        else {
                            disp as text "  p < " as res %5.3f 0.001
                        }

[Mel de Lange]

Thank you so much for your help George. When I run the code below I get the error: >0.001 invalid name.

Am I doing something wrong? (sorry am very new to Stata!)

levels VariableName
foreach i in `r(levels)'{
metan beta_ se_ if VariableName =="`i'", lcols(month)
if `pvalue'>0.001 {
disp as text " p = " as res %5.3f `pvalue'
}
else {
disp as text " p < " as res %5.3f 0.001
}
graph save "file_`i'", replace
}

[George Ford]

That code has to be inserted into the ado file. It won’t run within a do like that. You have to change the Metan code.

[Tiago Pereira]

I would approach this problem using gr_edit. The only challenge is to find the precise plot number on the graph, which can be accomplished through manual editing within the graph editor. I would proceed as follows:
a. run -metan- manually for one of the forest plots.
b. start the graph editor
c. start recording
b. edit the P-value manually
c. save the graph editing history - the file is a .grec
d. open the .grec file and check the correct plot number (for instance, plotregion1.plot22).

        clear
        set obs 10
        gene a = rbinomial(100,0.4)
        gene b = rbinomial(100,runiform()*0.3+0.1)
        gene c = rbinomial(100,0.4)
        gene d = rbinomial(100,0.4)
        
        metan a b c d, or random
        cap drop p1
        cap drop p2
        cap drop i2
        local p1 = r(p_het)
        gene p2 = `p1' in 1
        gene p3 = p2
        gene i2 = r(i_sq) in 1
        
        tostring p2, format(%9.2f) replace force
        local p2 = p2[1]
        
            
        tostring p3, format(%9.3f) replace force
        local p3 = p3[1]
                
        tostring i2, format(%9.1f) replace force
        local i2 = i[1]
        
        
        

            if `p1' < 0.001 {
            gr_edit .plotregion1.plot22.EditCustomLabel, j(01) label(I{superscript:2}=`i2'%,  Cochran's Q test, {it:P} < 0.001)
            }
            
            if `p1' >=0.01 {
            gr_edit .plotregion1.plot22.EditCustomLabel, j(01) label(I{superscript:2}=`i2'%, Cochran's Q test, {it:P} = `p2')
            }
            
            if `p1' <0.01 & `p1' >= 0.001 {
            gr_edit .plotregion1.plot22.EditCustomLabel, j(01) label(I{superscript:2}=`i2'%, Cochran's Q test, {it:P} = `p3')
            }

[George Ford]

Yup. If it's just the plot, then edit the graph.

[Mel de Lange]

Thanks so much George & Tiago. I will have a go at editing the graph!

[David Fisher]

Dear all,
As the co-author and maintainer of the metan package, I will look at implementing this within the next update.
Thanks,
David.