Difference-in-differences analysis with control variables

skclHA — Wed, 14 Feb 2024 05:35:22 GMT

HI, all.

I am currently running a DID analysis to examine the effects of policy interventions to depression.

I have a few questions about the control variables.

The current data format (example) is as follows.

Depression, age, income: continuous variables

Sex, work: dummy variables

pid	post	policy	depression	sex	age	income	work
1	0	1	16	1	50	200	1
1	1	1	15	1	50	300	1
2	0	0	20	0	30	500	1
2	1	0	35	0	30	400	0
3	0	0	17	0	20	600	0
3	1	0	50	0	20	900	1
4	0	1	35	1	40	800	0
4	1	1	25	1	40	400	0

1. In the above data format, should I time-variant control variables (income, work) convert a post-time value to a pre-time value (re-format 1) or a pre-time value convert to a post-time value (re-format 2)? Or should I use it in the current format?

RE-FORMAT (1)

pid	post	policy	depression	sex	age	income	Income_re	work	Work_re
1	0	1	16	1	50	200	200	1	1
1	1	1	15	1	50	300	200	1	1
2	0	0	20	0	30	500	500	1	1
2	1	0	35	0	30	400	500	0	1
3	0	0	17	0	20	600	600	0	0
3	1	0	50	0	20	900	600	1	0
4	0	1	35	1	40	800	800	0	0
4	1	1	25	1	40	400	800	0	0

RE-FORMAT (2)

pid	post	policy	depression	sex	age	income	Income_re	work	Work_re
1	0	1	16	1	50	200	300	1	1
1	1	1	15	1	50	300	300	1	1
2	0	0	20	0	30	500	400	1	0
2	1	0	35	0	30	400	400	0	0
3	0	0	17	0	20	600	900	0	1
3	1	0	50	0	20	900	900	1	1
4	0	1	35	1	40	800	400	0	0
4	1	1	25	1	40	400	400	0	0

2. If what I should use the data in current format is right, how should I make a syntax for the control variables (income, work) in the DID regression? And How should I interpret the results for dummy variables? I wrote currently syntax as bellow.

PROC MIXED DATA = LONG;

CLASS POST(REF="0") POLICY(REF="0") SEX(REF="0") WORK(REF="0");

MODEL DEPRESSION=POST|POLICY SEX AGE INCOME WORK / SOLUTION;

LSMEANS POST|EXPOSED / DIFF;

ESTIMATE 'D-I-D' EXPOSED*POST 1 -1 -1 1;

RANDOM Int/SUBJECT=PID TYPE=UN;

RUN;

3. Also, how would I write the syntax if I want to control for changes in the control variable (income, work) over time?

I apologize in advance for my poor English. Thanks!

Re: Difference-in-differences analysis with control variables

Ksharp — Wed, 14 Feb 2024 07:11:06 GMT

Maybe this could give you a help.
https://support.sas.com/kb/61/830.html

topic Difference-in-differences analysis with control variables in Statistical Procedures

Difference-in-differences analysis with control variables

Re: Difference-in-differences analysis with control variables