Understanding Card and Krueger (1994); Issues with T-test

[Maximiliano Rodriguez]

Hello everyone,

I have been reading Card and Krueger (1994) since I am going to develop a similar model for my thesis work. And first I wanted to replicate their work in order to understand what they did. When I was trying to do table 3, I got the following doubt. First, they got the mean value for FTE (full-time employment) before the policy change and the mean value for FTE after the policy change. This simply can be done with the following ttest command:

ttest emptot, by(pa) unequal
ttest emptot2, by(pa) unequal

Then I created the differences of these two variables:

gen d_emp=emptot2-emptot
ttest d_emp, by(pa) unequal

However, instead of getting row 3, I got row 3 of table 3 (Attached image). So, I tried to comparing the means of emptot2 and empto3 and I got the values of row 3, However, the standard errors are different, I think that they might have a different software of program back in that time.

by nj, sort: ttest emptot2=emptot, unpaired

What kind of command should I use in order to get row 3 of column 3 (iii) ? I tried running the diff command, but first reshaping the whole data with unsuccessful results.

Best Regards,

[Attaullah Shah]

I do not understand your question. I think you are already getting row 3 of table 3, with your command,

 
. ttest d_emp, by(pa) unequal

Two-sample t test with unequal variances
------------------------------------------------------------------------------
   Group |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
---------+--------------------------------------------------------------------
       0 |     309    .4666667    .4808286    8.452195   -.4794578    1.412791
       1 |      75   -2.283333    1.253269    10.85363   -4.780526    .2138597
---------+--------------------------------------------------------------------
combined |     384   -.0704427    .4604186    9.022325   -.9757072    .8348218
---------+--------------------------------------------------------------------
    diff |                2.75    1.342341                .0857844    5.414216
------------------------------------------------------------------------------
    diff = mean(0) - mean(1)                                      t =   2.0487
Ho: diff = 0                     Satterthwaite's degrees of freedom =  96.8838

    Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
 Pr(T < t) = 0.9784         Pr(|T| > |t|) = 0.0432          Pr(T > t) = 0.0216

[Maximiliano Rodriguez]

I do not understand your question. I think you are already getting row 3 of table 3, with your command,

. ttest d_emp, by(pa) unequal

Two-sample t test with unequal variances
------------------------------------------------------------------------------
Group | Obs Mean Std. Err. Std. Dev. [95% Conf. Interval]
---------+--------------------------------------------------------------------
0 | 309 .4666667 .4808286 8.452195 -.4794578 1.412791
1 | 75 -2.283333 1.253269 10.85363 -4.780526 .2138597
---------+--------------------------------------------------------------------
combined | 384 -.0704427 .4604186 9.022325 -.9757072 .8348218
---------+--------------------------------------------------------------------
diff | 2.75 1.342341 .0857844 5.414216
------------------------------------------------------------------------------
diff = mean(0) - mean(1) t = 2.0487
Ho: diff = 0 Satterthwaite's degrees of freedom = 96.8838

Ha: diff < 0 Ha: diff != 0 Ha: diff > 0
Pr(T < t) = 0.9784 Pr(|T| > |t|) = 0.0432 Pr(T > t) = 0.0216

Hi, Attaullah Shhah.

Actually, the results that you posted are for row 4 of Table 3. I got row 3 using reshape and then I used the diff command from (SSC) in Stata 14.2. The main difference is when I run the normal t test of the variable d_emp, the t test is conducted with a balance sample. The other method is to input the mean value per state (NJ and PA) of the variable emptot in wave 1 to those observations that have a missing value for emptot2.

In addition, if I used the reshape and then the diff command, I can replicate all the values of row 3 in table 3 presented in Card and Krueger (1994) something that cannot be done without using reshape command first.

Best Regards,