DID design

[Duong Le]

Hi all,

I am conducting a difference-in-difference study using four ways (2006, 2008, 2012, and 2014) of a repeated cross-sectional survey. In this study, I am estimating possible effects of a co-payment policy on older people's medical costs. The policy, which took effect in 2010, stipulates that those aged 80 and older or policy beneficiaries are entitled 100% free healthcare services so I have two possible treatment groups here. However, there also is another policy, which took place in 2003, stipulating that the poor is entitled 100% free healthcare services. I have several followings questions that need your suggestions.

1) Although the two treatment groups are different, their healthcare benefit are the same (100%) so does it make sense combine the two into one group?
2) I am using two-part model -twopm-, an user-written command by Belotti, Deb, Manning, and Norton to estimate outpatient costs. Before that, I want to test a parallel assumption trend using the codes below but I got an error message "no such variables"
where b_age is those aged 80 and older - treament group 1
policy stands for policy beneficiaries - treament group 2
year consists of 4 survey years
female (1=female; 0=male)
schoolyears means education in years
cost_op denotes out-patient costs
poorhh: poor household

* For those aged 80+
gen time=(year>2008)
gen age_did=b_age*time

reg cost_op i.b_age i.time i.age_did c.age c.schoolyears i.female if year<2012
     testparm  i.age_did

3) How to make a graph of the parallel assumption trend?
4) Marginal effects of the treated groups, are my codes below correct?

* Those aged 80+
twopm cost_op i.b_age i.time i.age_did c.age c.schoolyears i.female, f(logit) s(glm, family(gamma) link(log))
     margins if age_did==1, dydx(age_did)

5) Do I need to care for the possible effect of the 2003 policy? Or I just need to make sure that those who are entitled the 2010 policy are not the poor? Or Should I do a triple DID, accounting for the possible effect of the 2003 policy as well? In cases, if I want to do a triple DID so should the codes for a triple look like this?

gen triple=age_did*poorhh
twopm cost_op i.b_age i.time i.poorhh i.age_did i.b_age#i.poorhh i.time#i.poorhh i.triple c.age c.schoolyears i.female, f(logit) s(glm, family(gamma) link(log))
     margins if triple==1, dydx(triple)

Data

* Example generated by -dataex-. To install: ssc install dataex
clear
input float year int age float(b_age policy poorhh) byte female float schoolyears double cost_op
2014 61 0 0 0 1 12   400
2014 61 0 0 0 0 12   600
2014 60 0 0 0 0  9   150
2014 66 0 0 0 1 12     .
2014 78 0 0 0 1 12   540
2014 77 0 0 0 0  9    50
2014 73 0 0 0 1  9     .
2014 63 0 0 0 1 12  6550
2014 71 0 0 0 0 12     0
2014 66 0 0 0 0  9     .
2014 72 0 0 0 0 12   350
2014 93 1 0 0 0 12     0
2014 62 0 0 0 0 12  1000
2014 75 0 0 0 1 12  2500
2014 67 0 0 0 1  3   300
2014 77 0 0 0 1  5     .
2014 74 0 0 0 1 12 24000
2014 60 0 0 0 1 12   700
2014 65 0 0 0 1 12    10
2014 62 0 0 0 0 12     .
2014 62 0 0 0 1 12     0
2014 60 0 0 0 0  9     .
2014 75 0 0 0 0 12  1000
2014 65 0 1 0 0 12   600
2014 61 0 1 0 1  7  6000
2012 76 0 0 0 1  9     .
2012 72 0 0 0 0 12   600
2012 62 0 0 0 1 12     .
2012 66 0 0 0 1  9   750
2012 64 0 0 0 0 12     .
2012 72 0 0 0 0 12   600
2012 64 0 0 0 0 12     0
2012 60 0 0 0 1 12     .
2012 63 0 0 0 1 12     .
2012 73 0 0 0 1 12     .
2012 66 0 0 0 0 12     .
2012 74 0 0 0 0  6     .
2012 69 0 0 0 0 12     0
2012 66 0 0 0 0 12     .
2012 60 0 0 0 0 12     .
2012 76 0 0 0 1  3     .
2012 63 0 0 0 0  9  2000
2012 74 0 0 0 0 12     .
2012 64 0 0 0 0  9  1300
2012 60 0 0 0 0 10     .
2012 63 0 0 0 0 12   400
2012 60 0 0 0 0 12   200
2012 68 0 0 0 1 12   500
2012 69 0 0 0 0 12   120
2012 77 0 0 0 0 12     0
2008 60 0 0 0 0 12  2400
2008 65 0 0 0 0 12     .
2008 67 0 0 0 0 12 10150
2008 71 0 0 0 1 12  3600
2008 80 1 0 0 1  0   500
2008 67 0 0 0 0 12     0
2008 63 0 0 0 1 12   300
2008 84 1 0 0 1 12     .
2008 80 1 1 0 0 12   500
2008 60 0 0 0 0 12     .
2008 61 0 0 0 1 12     .
2008 61 0 0 0 1 12     .
2008 63 0 0 0 0 12     .
2008 64 0 0 0 0  9     0
2008 67 0 0 0 0 12     .
2008 83 0 0 1 1  0     .
2008 82 1 1 0 0  9  2500
2008 67 0 1 0 0  8   600
2008 63 0 1 0 0  9     .
2008 74 0 0 0 1  0  2400
2008 66 0 0 0 0  9  1500
2008 60 0 0 0 0  9   280
2008 63 0 0 0 1  2     .
2008 81 1 0 0 0  7     0
2008 60 0 1 0 0  9     .
2006 66 0 0 0 1  5     .
2006 66 0 0 0 0  9     .
2006 68 0 0 0 1  5     .
2006 64 0 0 0 0 12   300
2006 66 0 0 0 0  9   492
2006 63 0 0 0 1  2   570
2006 66 0 0 0 0 12   500
2006 79 0 0 0 1  3     .
2006 81 1 0 0 0 12     .
2006 66 0 0 0 0  0     .
2006 62 0 0 1 0  2     .
2006 69 0 0 0 1  9   560
2006 81 1 0 0 0  9     .
2006 69 0 0 1 0  0     .
2006 60 0 0 1 0 12     .
2006 67 0 0 0 0 12     .
2006 61 0 0 0 1  2     .
2006 60 0 0 0 1  6     .
2006 75 0 1 0 1  4  2000
2006 68 0 1 0 0  9     .
2006 85 1 0 0 1  0     .
2006 64 0 1 0 0 12   200
2006 67 0 0 0 1 12  5000
2006 62 0 1 0 1  9   300
2006 94 1 1 0 1  7     .
end
label values b_age b_age
label def b_age 0 "No", modify
label def b_age 1 "yes", modify
label values policy policy
label def policy 0 "No", modify
label def policy 1 "yes", modify
label values poorhh poorhh
label def poorhh 0 "No", modify
label def poorhh 1 "Yes", modify
label values female female
label def female 0 "male", modify
label def female 1 "female", modify

Any suggestion is highly appreciated!

Thank you.