capturing responses from the same individuals across waves

[Karen Arulsamy]

Hi all,

I would like to understand how 'dfruit_main' & 'dvege_main' (fruit & veg consumption) changed over time. These variables were only captured in particular waves in the survey and I would like to analyse responses in Waves 9,15 and 18.

How do I ensure that I am following the same individuals in these waves?

This is what I tried but I'm losing a lot of observations and the sample sizes are not constant across waves. It may not be surprising to lose some observations as Waves 15 & 18 were fielded during the pandemic.

gen surveyp = .
replace surveyp = 1 if wave == 9
replace surveyp = 2 if wave == 15
replace surveyp = 3 if wave == 18
tab surveyp

followed by

egen present = total(surveyp), by(pidp)
keep if present==3

----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. To install: ssc install dataex
clear
input long pidp float(surveyp dfruit_main dvege_main) byte wave
 22445 . . .  6
 22445 . 3 4  5
 22445 . . .  4
 22445 . 3 4  7
 22445 . . .  8
 22445 . 4 4 11
 22445 1 4 4  9
 22445 . . . 10
 29925 . . .  6
 29925 . . . 10
 29925 . . .  8
 29925 . 2 3 11
 29925 1 2 3  9
 29925 . . .  4
 29925 . 2 3  7
 76165 2 4 1 15
 76165 . . . 12
 76165 . . . 16
 76165 . . .  8
 76165 . . . 20
 76165 . 3 2 11
 76165 3 3 2 18
 76165 1 4 3  9
 76165 . 4 3  7
 76165 . . . 19
 76165 . . . 17
 76165 . . . 14
 76165 . . . 10
 76165 . . . 13
223725 . . .  7
223725 . . .  8
280165 . . .  8
280165 . 4 4  2
280165 . . . 14
280165 . . . 16
280165 . 4 4  7
280165 . . .  3
280165 . 4 4  5
280165 . 4 4 11
280165 . . . 12
280165 . . .  4
280165 . . .  6
280165 2 4 4 15
280165 . . . 10
280165 . . . 13
280165 1 3 4  9
333205 . 2 4 11
333205 . . . 10
333205 . . .  8
333205 . 4 4  7
333205 . . .  6
333205 1 3 4  9
387605 . 4 4  5
387605 . 3 2  7
387605 . . .  6
387605 . . .  4
469205 . . . 10
469205 1 2 2  9
469205 . . . 20
469205 . 2 2 11
469205 . . . 12
469205 . . . 16
541285 . . .  6
541285 . . .  3
541285 . . .  4
541285 . . .  5
541965 . . .  3
599765 . . . 20
599765 . . . 13
599765 . 4 4  5
599765 1 4 4  9
599765 . . . 14
599765 . . . 12
599765 . . .  4
599765 . 4 4 11
599765 . . . 10
665045 . . .  8
665045 . . .  4
665045 . 2 1 11
665045 . . .  6
665045 . . .  3
665045 . 2 2  5
665045 . . . 10
732365 . . . 16
732365 3 1 2 18
732365 . 2 1 11
732365 . . . 20
732365 1 2 3  9
732365 . . . 14
732365 . . . 10
732365 . . . 12
732365 . . . 19
732365 . . . 17
732365 . . . 13
760925 . . . 11
760925 . . . 10
813285 . . .  7
813285 . . .  4
813285 . . .  6
813285 . . .  8
end

Thanks in advance for your help with this.

Many thanks
Karen

[Fei Wang]

Karen, were you attempting to create a balanced panel data with only waves 9, 15, and 18? If yes, then,

keep if inlist(wave,9,15,18)
bys pidp: keep if _N == 3

In your data example, only "76165" is kept.

[Karen Arulsamy]

That's great, thanks so much for your help Fei Wang. In the main dataset I'm using, I have about ~10,000 observations per wave after keeping individuals who appear in all 3 waves.

[Karen Arulsamy]

Can I ask a follow up question, how can I ensure that the sample size is constant across waves for the outcome variables after using

keep if inlist(wave,9,15,18)
bys pidp: keep if _N == 3

In the example below, I would like to take the smallest sample size available in Wave 18 to get the tab results.

bysort wave: tab dfruit_main

> wave = 9

dfruit_main | Freq. Percent Cum.
------------+-----------------------------------
1 | 567 5.47 5.47
2 | 2,592 24.99 30.45
3 | 2,499 24.09 54.54
4 | 4,716 45.46 100.00
------------+-----------------------------------
Total | 10,374 100.00

---------------------------------------------------------------------------------------------------------------------------------------------------------------
-> wave = 15

dfruit_main | Freq. Percent Cum.
------------+-----------------------------------
1 | 562 5.45 5.45
2 | 2,726 26.44 31.89
3 | 2,133 20.68 52.57
4 | 4,891 47.43 100.00
------------+-----------------------------------
Total | 10,312 100.00

---------------------------------------------------------------------------------------------------------------------------------------------------------------
-> wave = 18

dfruit_main | Freq. Percent Cum.
------------+-----------------------------------
1 | 592 5.83 5.83
2 | 2,743 27.01 32.84
3 | 2,206 21.72 54.56
4 | 4,614 45.44 100.00
------------+-----------------------------------
Total | 10,155 100.00

* Example generated by -dataex-. To install: ssc install dataex
clear
input long pidp byte wave float(dfruit_main dvege_main)
   76165 15 4 1
   76165  9 4 3
   76165 18 3 2
 1587125 15 2 2
 1587125  9 3 3
 1587125 18 3 3
 4849085 18 4 4
 4849085 15 3 4
 4849085  9 2 2
68002725 18 2 2
68002725  9 2 3
68002725 15 2 2
68008847  9 3 2
68008847 15 3 3
68008847 18 2 2
68010887 18 2 3
68010887 15 4 4
68010887  9 3 4
68029931 15 2 2
68029931 18 2 2
68029931  9 2 2
68031967  9 2 3
68031967 15 2 2
68031967 18 3 2
68035365 15 4 2
68035365 18 2 2
68035365  9 4 3
68035367 18 3 3
68035367 15 3 3
68035367  9 4 3
68041487 18 3 3
68041487  9 4 3
68041487 15 3 3
68045567 15 4 4
68045567 18 4 4
68045567  9 4 4
68051007 18 2 2
68051007  9 2 3
68051007 15 2 2
68051011 18 4 4
68051011  9 4 4
68051011 15 4 4
68058487  9 4 4
68058487 18 4 4
68058487 15 4 4
68058491  9 4 4
68058491 15 4 4
68058491 18 4 3
68060531 15 2 3
68060531  9 2 3
end

Many thanks
Karen

[Nick Cox]

I don't understand #4. But harking back to #1: the total for any identifier if all 3 waves are present would be 6 = 1 + 2 + 3 not 3. That's why it didn't work as wanted.

[Karen Arulsamy]

Yikes, can't believe I missed that, thanks for pointing that out Nick.

With regards to #4 - after keeping individuals present in these three waves, the sample sizes are still different by waves so there may be item non-response. In regressions, I'd save the sample afterwards and use that sample throughout. I'd like to do this here so I can then see whether differences are due to changes in time or caused by differences in the sample. It might be obvious but I just can't seem to figure it out here.