Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Heckman selection model with difference-in-difference

    Good afternoon,

    I was advised to create a new thread on a previous one ( https://www.statalist.org/forums/for...interpretation).

    I will reiterate the background information- any help would be greatly appreciated.

    My topic regards a legislation's impact on disability in the workplace and concerns a 5 quarter longitudinal dataset whereby the legislation is enforced at the start of quarter 5. The use of logGRSSWK (the log of gross weekly wage- my dependent variable) only details information for 2 points: quarter 1 and quarter 5 (for which I acknowledge the analysis' profound limitations). Moreover, an important thing to also note is that a time trend was initially absent: the majority of variables ended in the number of the quarter they referred to (i.e GRSSWK1 = Gross weekly wage in quarter 1), although this was somewhat resolved via:
    Code:
     gen id = _n
    reshape long GRSSWK d_DIS DISCURR AGES URESMC REGWKR MARSTA HIQUAL8 HDPCH19 Inde07m GRSSWK2 TPEN091 STATR SKSBN91 FTPTWK , i(id) j(quarter)
    Nevertheless, I established a preliminary DID:
    HTML Code:
     reg logGRSSWK ACT DIS ACTDIS i.SEX i.ETH i.AGES i.RES_NEW i.REGWKR_NEW i.HDPCH19 i.IND i.MARSTA i.HIQUAL i.SKSBN91 i.FTPTWK, vce(robust)
    
    Linear regression                               Number of obs     =      3,836
                                                    F(74, 3761)       =      67.98
                                                    Prob > F          =     0.0000
                                                    R-squared         =     0.5729
                                                    Root MSE          =     .53276
    
    --------------------------------------------------------------------------------------------------------
                                           |               Robust
                                 logGRSSWK |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
    ---------------------------------------+----------------------------------------------------------------
                                       ACT |   .1165873   .0193822     6.02   0.000     .0785866    .1545879
                                       DIS |   -.033673   .0392651    -0.86   0.391    -.1106559      .04331
                                    ACTDIS |  -.0090192   .0517852    -0.17   0.862     -.110549    .0925106
    (where DIS= 1 if disabled, 0 if non-disabled; ACT = 1 if post-legislation, 0 if otherwise; ACTDIS = ACT *DIS)

    However, this restricts analysis to those in employment and, to ensure the pre- and post- compositions didn't change to undertake the DID, I ensured all observations were in employment quarters 1 and 5 (further limiting the validity of results).

    Therefore, my main question is: is the Heckman selection model still applicable given this limited background and, if so, does the required code vary due to this context?

    Any help would be greatly appreciated, thank you.
Working...
X