Homework 4 Brainstorm

docx

School

Rutgers University *

*We aren’t endorsed by this school

Course

312

Subject

Statistics

Date

Feb 20, 2024

Type

docx

Pages

Uploaded by BailiffMorning7702

Homework 4 Brainstorm 1. What to Report: a. Provide a list of all the variables you intend to consider as confounding covariates by group (most important to balance, somewhat important to balance, and least important to balance) More important to balance: momrace momage momed brthwt Somewhat important to balance: Hispanic black white hs ltcoll college Least important to balance: brordum b_marr pr0 lths preterm b. In 4 or fewer sentences describe why this set of variables was chosen and how they were classified into groups. 2. What to report: a. Provide one set of balance statistics (for the model of choice). b. Report (in tabular form if possible) how many variables in each of your priority categories (from question 1) meet the balance targets. c. Cut and past the output from psbal2 from your model of choice [SB This might be the same as B] 3. Plot and Interpret the overlap histograms from your model to determine if the counterfactual units available are sufficient for the entire treatment group. Remember this plot shows distributions pre-match. Use no more than 3 sentences. Be sure to include this plot in the document you turn in. 4. Using your model from (2), answer the following: a. Calculate ( and report ) a regression-adjusted treatment effect estimate (with confounding covariates included in the regression) and corresponding standard error. Cut and paste your Stata command and relevant output and highlight the appropriate estimates. b. Interpret the treatment effect estimate (even if it is not statistically significant) from (b) causally ; use no more than two sentences. Make sure to use causal language and to be clear about the counterfactual state and the group about whom we are making inferences. c. Create a histogram/response surface plot (include this plot in the document you turn in). What does this plot suggest about the potential heterogeneity in the treatment effect? How might we expect the effect of the treatment on the treated to differ from the effect of the treatment on the controls? Answer these questions using no more than two sentences. d. Describe what the four most important assumptions are to interpret the estimates in (a) and (b) causally? Assess the plausibility of each. 5. Suppose I repeated the process above ((1) – (4)) but instead used the variable not_first rather than first in the psmatch2 command in (2), where “not_first” is equal to 1-first .

All other commands would remain the same (that is the regression would include first rather than not_first ). What estimand would the analysis in 4b be estimating…? (Recall here the regression equation includes first , rather than not_first ) Answer using one sentence or less. 6. Suppose I ran a regression of the outcome on the treatment indicator with probability weights equal to 1/pscore for the people in the treatment ( first=1 ) group and 1/(1- pscore) for those in the control ( not_first=1 ) [where pscore was calculated using first as the dependent variable]. What causal estimand is this approach trying to estimate? Answer using one sentence. 7. Perform a standard regression (no matching) of piatm56 on first and all variables you considered to be confounding covariates in building your model. a. Report the estimate (again, cut and paste computer output ) and this time interpret it without making a causal statement (i.e. interpret it descriptively ). Answer using one sentence. b. How would the assumptions needed to interpret the estimate from this analysis causally differ from the propensity score matching analysis? No more than 4 sentences. 8. The analyses in questions 4-6, besides using slightly different methodological strategies, also estimate different quantities, or estimands. Which estimand is the most policy-relevant from your perspective. Describe the estimand and describe a real or hypothetical policy that it could inform using 3 or 4 sentences. [Please be clear that I am asking for an estimand not an estimate. ]

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version