Notice: On March 31, it was **announced** that Statalist is moving from an email list to a **forum**. The old list will shut down on April 23, and its replacement, **statalist.org** is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
Jeremy Wells <jwell33@tigers.lsu.edu> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: Re: st: xtlogit - lagged dependent variable as independent variable |

Date |
Tue, 23 Apr 2013 09:17:25 -0500 |

Andreas, I am not a methodologist, so please do not take my advice as any sort of authoritative position. I have dealt with these issues quite a bit in my own work though, so I have some personal preferences. Again, I would strongly recommend Beck and Katz's work. In the article I referenced, they conclude, "there is nothing about lagged dependent variables that makes them generically harmful" (350-351). They also conclude that adding a second period lag (L2.DV plus lags of all independent variables, which they call ADLLDV2) is actually preferred, though they do not go so far as to have four lags. Again, this is the conclusion of Beck and Katz, and they have several critics. Also, they do not mention the Arellano-Bond method, and I am not familiar with it, so I cannot offer any help there. In terms of your specific case, I would be concerned about including a four period lag with only ten years worth of data. You would essentially be throwing out almost the first half of your data. Just keep that in mind. Jeremy Wells Ph.D. Student LSU Dept. of Poli. Sci. 324 Stubbs Hall Baton Rouge, LA 70803 Jeremy, thanks a lot for your fast and detailed answer. One short question just to make sure: As I understand from your answer there is no further adjustment - comparable to the Arellano-Bond adjustment - needed when using a lagged dependent variable (given my dataset characteristics and the fact that I use -xtlogit, re- and not a linear regression model). Does this stand true if I add more than one lagged variable? -xtlogit strategy L.strategy L2.strategy L3.strategy L4.strategy company_age company_size industry, re- I want to use this model to check how many years back have an influence on the choice of the strategy in the current period. Thanks a lot and best wishes Andreas Schiffelholz Am 22.04.2013 22:51, schrieb Jeremy Wells: Andreas, There is a pretty intense debate over how to model autocorrelation, especially within political science. I would suggest work by Beck and Katz, especially in the Annual Review of Political Science (2011, vol. 14: 331-352). They prefer lagging the dependent variable, but there are other methods. I would also suggest using the time-series operators built into Stata to lag your DV, so your command would look like this: -xtlogit strategy L.strategy company_age company_size industry, re- (I also took out the year variable, because I am not sure how well that would factor in with a lagged DV and the company age variable, though it may not be problematic or they may be unrelated, but I would bet there would be a lot of multicollinearity there.) HTH. Jeremy Wells Ph.D. Student LSU Dept. of Poli. Sci. 324 Stubbs Hall Baton Rouge, LA 70803 Sorry, the Stata command mentioned below got messed up: - xtlogit strategy strategy_lag1 company_age company_size industry year, re - Andreas Am 22.04.2013 20:15, schrieb Andreas Schiffelholz: Hello, I'm currently working with a company dataset in the form of an unbalanced panel (overall sample size: 1.300 company years, T: 10, X: ~170 different companies). One of two strategic types was assigned to each of the company years. To get a better understanding, why companies are pursuing a specific strategic type, I am using a random effects model -xtlogit, re-: - xtlogitstrategystrategy_lag1company_agecompany_sizeindustryyear, re - with "industry" and "year" being a set of dummy variables. Part of this model is the independent variable "strategy_lag1" which is the strategic type (dependent variable) of the previous year. This variable is added to get a better understanding of the stability of the strategic type in terms of time. As far as I know there is an adjustment of the model needed when adding lagged dependent variables to the model. I did find the -xtabond- command for linear models, which is using the adjustment procedure suggested by Arellano, Bond (1991). For the xtlogit model I did not find a comparable command. Given the characteristics of my data set, is there an adjustment of the standard -xtlogit, re- model needed? Does the answer to this question change if I add additional variables with longer lags to the regression (strategy_lag2, strategy_lag3, …)? If so, is this adjustment implemented into Stata or does anybody know a user programmed command dealing with this issue? Thanks very much, Andreas Schiffelholz P.S.: This is my first time posting something on Statalist. If the description of my problem is not precise enough or if I broke a specific rule of the list, please let me know. * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**st: New command -scoregrp-** - Next by Date:
**st: xtivreg2 testing for regressor endogeneity** - Previous by thread:
**Re: st: xtlogit - lagged dependent variable as independent variable** - Next by thread:
**st: mata orgtype** - Index(es):