Manipulation Of Data

1206 Words3 Pages

Manipulation of Data Before running the regressions in STATA, the data was modified for missing observations. These were originally coded as 99999 for missing values, which would generate very incorrect coefficients and so they were replaced with a “.” to omit them from the regression. Difference in Difference For the simple difference in difference I am regressing the interaction term of post along with the fixed effect of periods and ever_treat on the dependent variable weeks to observe how these control variables affected the total number of weeks addicts worked when they were being treated compared to addicts who were not treated. Simple Difference in Difference equation: weeks_it=β_0+β_1 ever_treat_i+〖δ_t+β〗_2 post_i+u_i Explanation of variables in Equation In the equation the control ever_treat means whether or not one has been part of treatment group. It is the equivalent of the treatment group in a Difference in Difference and is used as dummy variable to separate whether the individual had ever been treated, regardless of period. It is assigned a value of one if the addict was ever treated and zero if the addict was never treated. 〖Futhermore,δ〗_t denotes the time effects or period fixed effects that accounts for overall changes between periods. This time period allows for the analysis of the means during a given time period; however, period 1 is lost due to fixed effects, so all period coefficients are in comparison to period 1 which is omitted. The variable post is the equivalent of interaction term between post and treatment and captures the effect of the treatment on the amount of weeks worked. Data Table of Regression weeks_it=β_0+β_1 ever_treat_i+〖δ_t+β〗_2 post_i+u_i (1) VARIABLES weeks ... ... middle of paper ... ...observed in the data because when analyzing the means of the variables there exists an inherit difference in the levels of education between those in the control group vs. those who were treated. In the control group the mean education level was 10.8 years while those addicts who were treated had a mean education level of 12.1 years. As can be seen in the previous table the coefficient of educ is 0.715, which is positive, and so has a positive effect on the amount of weeks worked by addicts who were in the treatment program. This difference in education levels could have compounded the effect of the treatment (post) making it seem more effective than it actually was. As Displayed in the following tables the majority of the variables’ means were extremely similar and so unbiased, yet the one variable that displays bias in favor of the treatment group is education.

Open Document