Soc 382 Event History HW4 Rev 2/19/2019
Consider both the single observation dataset (dataset1) and the couple-month dataset (dataset 3) available on Canvas.
1) Explain the concepts of left and right censoring, and how they relate to the HCMST study of predictors of breakup we are using in HW 4. Why might left censoring and right censoring each be a problem? How does Event History Analysis deal with left and right censoring?
2) Run the command stset on both dataset 1 and dataset 3, to reveal the existing stset for both datasets, and explain as much as you can about the items in the stset.
3) Using dataset 3, run the following command that generates a Kaplan-Meier couple survival curve:
sts graph, by(married_tv)
sts list, by(married_tv) at(0 1 2 3 4 5 10 15 20 25 30)
The sts list function will give precise values and SE for the survivor function at each year of couple duration. Comment on the shape of the curves from the Kaplan-Meier graph, and comment on the small number of marriage failures recorded in the first 5 years of relationship duration (note: relationship duration, not marital duration).
4) Run the following model on dataset 3, and explain the hazard ratio output. What factors most significantly predict breakup, and why?
stcox resp_col_dgre_tv married_tv coresident_tv children_in_hh_tv met_online_augmented ln_hhinc_2009dollars_tv relationship_quality_rescaled ppage
5a) Run the discrete time event history analog to the above Cox proportional hazards model using dataset 3, and the logistic command (or logit with the “or” option) to get odds ratio output. Use broke_up_tv_v2 as the dependent variable, and how_long_relationship_tv as the term controlling for relationship duration. Compare to the Cox model in Q4, and explain the similarities or differences.
5b) Run the same discrete time event history model as 5a, but add in inv_relationship_duration_tv as a second term explaining relationship duration’s effect on breakup. Does model 5b fit better than 5a by the Likelihood Ratio Test? How do 5a and 5b compare by the BIC (use n=2649, the number of couples with nonmissing data on relationship duration).
6) Run an analog (with the same variables, in their non-time-covarying form) of the model from Q4 on the single observation dataset. Comment on its similarity and difference (in outcomes and assumptions) with the model in Q4.