Controlling for both level and difference in variables between partners

[Chris Boulis]

Using panel data, I'd like to measure the effect of life satisfaction in couples on generated wealth. To date, I've been using a variable that measures the difference in the level of life satisfaction between partners in which I take the absolute difference

gen byte satdiff = abs(losat1 - losat2)

Where 'losat' ranges from 0 (completely dissatisfied) to 5 (neither disatisfied/satisfied) to 10 (completely satisfied) and '1' refers to male, '2' refers to female.

However, the difference variable does not take account of couples in which both parters were completely dissatisfied so I want to take account of the effect of both the level and difference in life satisfaction in couples. I have tried including both the level and difference variables in the regression (c.losat1 c.losat2 and satdiff), but in doing so, the difference variable is no longer significant - potentially due to being correlated with the level variables. Ideas appreciated.

xtreg c.wealth2r i.at3 c.hgage1 c.hgage2 c.satdiff c.lincr

Data sample:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float wealth2r byte at3 int(hgage1 hgage2) byte(losat1 losat2 satdiff) float incr
1070211.4 0 37 38  6  9 3  100396.3
 877808.4 0 55 41  6  6 0 101745.04
 20581.24 0 22 19  8  8 0 11889.035
1235545.5 0 57 55  9  7 2  163804.5
 706076.6 0 34 33  8  9 1  20475.56
  43163.8 0 36 34  8  6 2  46235.14
 231907.5 0 52 50  5  7 2  40026.42
241455.73 0 45 47  7  7 0 154198.14
 428661.8 0 44 50  9  6 3 123117.56
 85898.28 0 35 34  6  5 1  64729.19
 5200.792 0 30 27  8  7 1  50858.65
137582.56 0 34 33  9  8 1  81638.04
 813474.2 0 57 54  9  6 3  138705.4
 170805.8 0 26 23  8  9 1 121532.36
 567063.4 0 49 44  9  9 0 127381.77
 44696.17 0 37 28  3  5 2  51630.12
  2583231 0 60 53  9  9 0 116248.34
 99077.94 0 37 31  6  7 1  85865.26
2716578.5 0 40 36  7  8 1  524001.3
131933.94 0 27 27  8  8 0 123972.26
 98414.79 0 46 39  6  4 2  57595.77
 94433.29 0 32 30  7  9 2  94731.84
 98857.33 0 21 21  8  9 1 16512.549
 415455.7 0 41 35  9 10 1  37384.41
end

[Carlo Lazzaro]

Chris:
why not considering the issue from the opposite side, that is in additive instead of subtractive terms:

 g try_it= losat1 + losat2

This way, you create a categorical predictor that allows -los1=0- and -los2=0- in the same couple.

[Chris Boulis]

Thank you Carlo Lazzaro. That's a very good idea and removes the need to include 'losat' separately for each partner.

I believe there is also identifying the effect of differences as it is expected to give rise to conflict in relationships. I tried creating a variable with up to three-levels of difference but Stata advises my syntax is invalid - any suggestions how I can address this?

gen byte satdiff1 = 0 if abs(losat1 - losat2) = 0 // same level  
replace satdiff1 = 1 if abs(losat1 - losat2) = 1 // 1 level diff
replace satdiff1 = 2 if abs(losat1 - losat2) = 2 // 2 level diff
replace satdiff1 = 3 if abs(losat1 - losat2) = 3 // 3 level diff

Stata v.15.1. Running panel data.

[Carlo Lazzaro]

Chris:
it might be an issue of a missing double = sign after the -if- clause:

. use "C:\Program Files\Stata17\ado\base\a\auto.dta"
(1978 automobile data)


. gen price2=1 if (price-mpg)==100
(74 missing values generated)

. gen price3=1 if (price-mpg)=100
invalid syntax
r(198);

.

[Chris Boulis]

Yes, that's it - I can't believe I didn't see that. Thank you Carlo Lazzaro.