I have panel data of companies from various industries for 10 years. However, my data set contain many outliers. One of the way that I know to remove them is using bloxspot to identify outliers and den delete them manually. However, it takes enough time. I would like to know syntax preferably in STATA to remove such outliers with one command. Thanks.
Hi Umar,
I am not a fan of removing outliers, as they typically carry meaning. That said, here are some options for you to consider. winsor is a user-written program that allows you identify outliers and replace their values with those below the cutoff. grubbs is a user-written program that identifies outliers using the Grubbs (1969) approach. bacon is a user-written program for identifying multivariate outliers. All of these programs can be downloaded into Stata by typing: ssc install program (where program is the name of the program I listed above)
I hope this helps
Ariel
Hi Umar,
I am not a fan of removing outliers, as they typically carry meaning. That said, here are some options for you to consider. winsor is a user-written program that allows you identify outliers and replace their values with those below the cutoff. grubbs is a user-written program that identifies outliers using the Grubbs (1969) approach. bacon is a user-written program for identifying multivariate outliers. All of these programs can be downloaded into Stata by typing: ssc install program (where program is the name of the program I listed above)
I hope this helps
Ariel
Tuan,
Outliers do not bias the results if they represent true observations. They only make it more difficult to determine the appropriate functional form of the model. They can bias the results if they do not represent true observations (e.g. data entry error).
I am also not clear as to why you set the cutoff at the 99th percentile? What if outliers go as deep as the 96th percentile or if there is only one outlier that is in the 99.9th percentile? This arbitrary cutoff may impact the validity of the study more than if you leave the outlier in the data.
Ariel
You may also want to consider a quantile regression (qreg in stata) which is robust to outliers. Alternatively, you may find "Robust Regression in Stata" (http://www.stata-journal.com/sjpdf.html?articlenum=st0173) interesting.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software