dtable difference between dimensions var and colname plus sample frequency not displayed when combining collections

[Suzanna Vidmar]

I thought I wouldn't have any further questions about dtable and collect for the moment but have two more. I'm sorry to be taking up so much time and space on Statalist of late. I hope other Stata users are get some benefit from my posts.

Here's the code and the output along the way.

*_______________________________________________________________________________
*Creating the example dataset.

*Reading in the dataset.
use https://www.stata-press.com/data/r18/exercise.dta,clear
keep if inlist(day, 0, 4, 6, 8)

*Replacing the variable program with group_allocation, where 1=intervention and 2=control.
gen group_allocation=program
lab var group_allocation "Group allocation"
lab def grplbl 1 Intervention 2 Control
lab val group_allocation grplbl
drop program

gen blinding=2
replace blinding=3 if _n/3==int(_n/3)
bysort day: replace blinding=1 if _n/4==int(_n/4) & day!=0

lab def blindlbl 1 "Change or lack of change from baseline" ///
                 2  "Participant informed assessor" ///
                 3  "Guess"
lab val blinding blindlbl


*_______________________________________________________________________________
*Creating the table.

*Clearing any collections that may be in Stata's memory.
collect clear

*Creating a separate descriptive table for each level of time.
*For time=1, including the sample frequency in the column header.    
dtable i.blinding if day==0, ///
    by(group_allocation,nototal) ///
    sample(, statistic(frequency) place(seplabels)) ///
    sformat("n=%s" frequency) ///
    title(Table 5: Blinding of assessor) ///
    titlestyles(font(Calibri, size(12) bold)) ///
    name(Day0)

forval i=4(2)8 {
    dtable i.blinding if day==`i', ///
        by(group_allocation,nototal) ///
        nosample name(Day`i')
}


*Combining the above collections.
collect combine comb = Day0 Day4 Day6 Day8
collect query composite _dtable_stats

OUTPUT:
Composite definition
Collection: comb
 Composite: _dtable_stats
  Elements: frequency
            fvfrequency
            fvpercent
 Delimiter: " "
      Trim: on
  Override: off

The sample frequency is in _dtable_stats. The following command also confirms this.

. collect label list result

  Collection: comb
   Dimension: result
       Label: Result
Level labels:
   frequency  Frequency
 fvfrequency  Factor-variable frequency
   fvpercent  Factor-variable percent

Now for the rest of the code and the table produced.

*Setting the cell percentages to integers.
collect style cell result[fvpercent], nformat(%3.0f)

*Adding a border to separate each level of time
collect style cell colname[3.blinding], border( bottom, width(1))

*Creating the final table.
collect layout (collection#colname) (group_allocation#result)

OUTPUT:

Table 5: Blinding of assessor
----------------------------------------------------------------
                                              Group allocation  
                                           Intervention  Control
----------------------------------------------------------------
Day0                                                            
  blinding                                                      
    Participant informed assessor              11 (69%) 14 (67%)
    Guess                                       5 (31%)  7 (33%)
----------------------------------------------------------------
Day4                                                            
  blinding                                                      
    Change or lack of change from baseline      4 (25%)  5 (24%)
    Participant informed assessor               8 (50%) 12 (57%)
    Guess                                       4 (25%)  4 (19%)
----------------------------------------------------------------
Day6                                                            
  blinding                                                      
    Change or lack of change from baseline      2 (12%)  7 (33%)
    Participant informed assessor               9 (56%) 10 (48%)
    Guess                                       5 (31%)  4 (19%)
----------------------------------------------------------------
Day8                                                            
  blinding                                                      
    Change or lack of change from baseline      4 (25%)  5 (24%)
    Participant informed assessor               8 (50%) 10 (48%)
    Guess                                       4 (25%)  6 (29%)
----------------------------------------------------------------

Question 1: How do I display the sample frequency in the header for the Intervention and Control groups?

Question 2: What is the difference between the dimensions colname and var? When I use var instead of colname for the row component, results are shown for only 2 levels of my variable blinding. There genuinely are only 2 levels for Day=0 but there are 3 levels for the other days.

. collect layout (collection#var) (group_allocation#result)

Collection: comb
      Rows: collection#var
   Columns: group_allocation#result
   Table 1: 16 x 2

Table 5: Blinding of assessor
-------------------------------------------------------
                                     Group allocation  
                                  Intervention  Control
-------------------------------------------------------
Day0                                                   
  blinding                                             
    Participant informed assessor     11 (69%) 14 (67%)
    Guess                              5 (31%)  7 (33%)
Day4                                                   
  blinding                                             
    Participant informed assessor      8 (50%) 12 (57%)
    Guess                              4 (25%)  4 (19%)
Day6                                                   
  blinding                                             
    Participant informed assessor      9 (56%) 10 (48%)
    Guess                              5 (31%)  4 (19%)
Day8                                                   
  blinding                                             
    Participant informed assessor      8 (50%) 10 (48%)
    Guess                              4 (25%)  6 (29%)
-------------------------------------------------------

Kind regards,
Suzanna

[Jeff Pitblado (StataCorp)]

2. The auto levels style for var is set in collection Day0, and takes precedence in collection comb. This is why there is a missing level. The auto levels style for colname is not set by dtable. Clear the auto levels style for var to get the same behavior as with colname.

collect style autolevels var, clear

1. If you look at the layout specification in collection Day0, which was constructed using sample() sub-option place(seplabels), you will see that the column specification includes dimension _dtable_sample_dim which is how dtable gets the sample sizes into the column heading. To get this column heading in your combined table, you need to add _dtable_sample_dim tags in the other collections before you combine them, then add _dtable_sample_dim to your column specification in your layout.

To accomplish this, I changed the loop creating the other collections to

forval i=4(2)8 {
    dtable i.blinding if day==`i', ///
        by(group_allocation,nototal) ///
        nosample name(Day`i')
    collect addtags _dtable_sample_dim[1], fortags(group_allocation[1])
    collect addtags _dtable_sample_dim[2], fortags(group_allocation[2])
}

and I changed the layout to

collect layout (collection#colname) (group_allocation#result#_dtable_sample_dim)

or

collect style autolevels var, clear
collect layout (collection#var) (group_allocation#result#_dtable_sample_dim)

[Suzanna Vidmar]

Thank-you Jeff. That works beautifully.

Cheers,
Suzanna