Hello,
I am uncertain about what exact methodology I should be using to test my hypothesis. I am looking at determinants of exit strategies for businesses. In particular, I want to see whether governance related variables such as the generation in control or the presence of an outside manager influence the likelihood of three different exit strategies. Now, I am puzzled by what exact model to use.
The exit event is recorded in yearly intervals, so I assume I need to use a model for discrete data. Additionally, the three different exits are "competing risks", meaning that firms are excluded from the sample after they have followed one of the exit strategies, so a firm cannot at the same time follow two of the exit strategies. I do have the incorporation date of the firm, so no left-censoring is present. However, I have right-censoring as the dataset only follows the firms until 2015 and a lot of them have not exited.
I found a paper by Janitza & Tutz (2015), in which state that the standard approach for time discrete competing risks "consists of fitting a multinominal logit model which links an individual's covariates to the risk for observing a specific event". However, I did find other papers that split up the competing risks into separate groups and used a cox proportional hazard model.
I appreciate any help very much as I do not know which method is the appropriate. Modeling everything in Stata should be doable with Stephen Jenkins' web lessons I guess.
Best wishes, Tom
I am uncertain about what exact methodology I should be using to test my hypothesis. I am looking at determinants of exit strategies for businesses. In particular, I want to see whether governance related variables such as the generation in control or the presence of an outside manager influence the likelihood of three different exit strategies. Now, I am puzzled by what exact model to use.
The exit event is recorded in yearly intervals, so I assume I need to use a model for discrete data. Additionally, the three different exits are "competing risks", meaning that firms are excluded from the sample after they have followed one of the exit strategies, so a firm cannot at the same time follow two of the exit strategies. I do have the incorporation date of the firm, so no left-censoring is present. However, I have right-censoring as the dataset only follows the firms until 2015 and a lot of them have not exited.
I found a paper by Janitza & Tutz (2015), in which state that the standard approach for time discrete competing risks "consists of fitting a multinominal logit model which links an individual's covariates to the risk for observing a specific event". However, I did find other papers that split up the competing risks into separate groups and used a cox proportional hazard model.
I appreciate any help very much as I do not know which method is the appropriate. Modeling everything in Stata should be doable with Stephen Jenkins' web lessons I guess.
Best wishes, Tom
