Anda di halaman 1dari 3

Sessions 1-7 - Lab assignment This assignment will be done over two sessions, and you will continue

building on this dofile in subsequent sessions, so save your dofile (on a USB, or email it to yourself). Email your wor so far on the dofile to the T! before leaving lab at the end of the session. "a e sure your first and last names are clearly indicated. #ur ob$ective is to test whether im%roved toilet facilities affect child health as measured by the weight for age &'score, using the "()S *++,'- and *+.. datasets. /e are %re%aring data to run the following regression0 1i 2 + 3 . T#(4ETi 3 * /E!4T5 3 ei /here0 1 is the weight'for'age 1'score of the child T#(4ET is a dummy for having im%roved toilet facilities /E!4T5 is a household wealth inde6 (already calculated by 7un$ab Bureau of Statistics) 8o all the following ste%s in a dofile. /here you need to comment on the res%onses to the questions, annotate the dofile by %receding the relevant lines with an asteris (9). .) #%en "()S *+.. household data file. *) :ind the geogra%hic and 55 (8s that uniquely identify the household. ;) (dentify and summari&e the ey variables we will need to run our regression. <) Tabulate the toilet ty%e variable, with and without labels. 8ecide how you want to define =im%roved toilet>. )reate a new variable, toilet, which is . for an im%roved toilet, and + otherwise. 8o you need to re%lace any values as missing? /hy @ why not? A) 5istogram the weight'for'age 1'score variable. 8o you need to re%lace any values as missing? /hy @ why not? B) )reate a variable, round, to denote the round of the survey. ,) "erge household datafile with child datafile. /hich of the two has unique values? -) Ceview the data using summari&e and browse commands and chec that everything has wor ed %ro%erly. 7ay s%ecial attention to the Dmerge variable. /hat does each value re%resent? /hy are there some values of . in this variable? E) #%en "()S *++,'- child'household data. !re all the variables %resent that are required for the regression? )hange any variable names as needed so they match the "()S *+.. dataset. .+) )reate a round variable in both rounds of the survey so that when you a%%end the two it is clear which observations are from which round. ..) !%%end the * survey rounds. .*) Ceview the data using summari&e and browse commands and chec that everything has wor ed %ro%erly. .;) Cun the regression s%ecification above (this is a %ooled cross section regression).

.<) "a e a table to dis%lay the results using the outreg command. .A) !t what level do you thin the standard errors should be clustered? /hy? Cun the regression again with the a%%ro%riate clustering, using the synta6 reg y 6, cluster(cluster(8) .B) !dd the results of the regression with clustered SE to the table ne6t to the original results. 5ow do the results differ? .,) /hat inds of #FB could occur in this regression? /hat variables in the dataset could you include as controls to hel% address this? Browse variable labels and@or the questionnaire (on the 8ro%bo6) to identify a%%ro%riate variables. .-) 4oo at the descri%tives and distribution of these new control variables and assess whether you need to do any cleaning of these variables, as you did with the main study variables above. 7erform any needed cleaning. .E) Cun the regression again with clustering and the additional controls, and %ut it in the table ne6t to the first two columns. 5ow do the results differ? *+) Gow consider that you are designing a new study on 7un$ab. Hou want to test whether two biradaris have differences in wealth, as measured by the number of cattle owned. Gote that the "()S survey does not include information on biradari, so you need to design a new sam%le survey which does as for this information. Use the information from the "()S and StataIs sam%si command to calculate the a%%ro%riate sam%le si&e. S%ecify what effect si&e, al%ha and beta you are using and what each of them means. *.) 4oad the %smatch* command and use it to carry out a %ro%ensity'score matching a%%roach to the same research question as the one in the regression above. /hat variables will you %ut in the %ro%ensity score equation? %robit treatedvar 6. 6* 6; %redict %score %smatch* treatedvar, outcome(outcomevar) %score(%score) neighbor(.) common Gote0 neighbor(.) can be re%laced or augmented with other o%tions to change the matching method. Gote0 the out%ut com%ares the unmatched results (i.e. without using 7S") to the !TT (!verage Treatment effect on the Treated) estimated by 7S". **) /hat was the range of common su%%ort? *;) Use %stest %ost estimation command to chec for balancedness. (nter%ret the results. *<) !dd these results to the table you made earlier using #4S. 5ow and why do these results differ from the results of the #4S models? *A) Gow we will use tehsil'level averages as an instrument for whether an individual household has an im%roved toilet. )onstruct a tehsil'level average for the T#(4ET dummy. *B) Cun *S4S, using the tehsil average as an instrument. )ontinue to use clustered standard errors in this estimate as before. *,) (s the instrument informative? /hat test statistic do you need to use here? Ce%ort its value and inter%ret. *-) (s the instrument valid? E6%lain what as%ects of endogeneity it addresses. Jive one e6am%le of a %ossible violation of the instrumentIs validity.

*E) )an you use the overid test here? /hy or why not? ;+) Gow use tehsil average and tehsil average squared as instruments. Gow can you use the overid test? ;.) Cun the overid test, re%ort the results and inter%ret. ;*) Cun the 5ausman test, re%ort the results and inter%ret. ;;) Cun the original regression with district fi6ed effects, re%ort the results and inter%ret. ;<) Cun the original regression with tehsil fi6ed effects, re%ort the results and inter%ret.

Anda mungkin juga menyukai