Hello
I have two questions:
1. Can I use more than one method for filling the missing values? Like moving averages, and imputing mean where moving averages didn't work?
2. Are the methods for treating missing data in panel data same like any other? Can I use 2 period moving averages if when I plot it on excel '2 period moving averages' is the best and closest representation of my available datapoints?
Thanks.
Currently, there are three specific strategies for handling missing data: (1) imputation of the missing values with suitable estimates, (2) analysis of incomplete data without adjustment, and (3) discarding of nonrespondents and calculation of weights for respondents to compensate for the loss of cases. The merits and demerits of these approaches must be weighed carefully before choosing which one to adopt. The second broad recommendation is the practice of partially following up on hard-to-interview respondents. For example, a researcher who has lost 150 of 1000 subjects by a later wave may wish to attempt locating all 150 missing subjects, but the outcome of this costly and time-consuming endeavor is unlikely to yield a 100% response rate. Instead, the researcher should devote resources to tracking a random sample of those 150, and then obtain valid survey-based information for those individuals. With this information, the researcher could create confidence intervals or boundaries around the estimates of interest.
With regard to mortality, perhaps the best approach is to consult the National Death Register Index, which contains information on the individual's name, date of death, and cause of death. Another approach would be to search the obituaries of newspapers. Implementation of both strategies is recommended. In terms of exposure time, two approaches are worthwhile. Regarding official data sources, researchers could obtain exposure time information regarding hospital stays and/or prison/jail stays from officially kept records at the federal, state, or local level. Because some of this information is inadequate, especially jail records, which oftentimes do not contain movement information (i.e., from jail to probation, number of days in jail, etc.), researchers could ask respondents to self-report information regarding exposure time.
Hello,
Yes, you can use multiple methods for filling in missing values in panel data, such as combining moving averages with mean imputation. The methods for handling missing data in panel data are similar to those in other data types, and using 2-period moving averages is valid if it closely represents your available data points.
Hope this helps
Hi Tina, I always thought mean substitution was used, but I recall sometime ago reading someone's answer on this site, for another question and I can't recall the exact question or the person whom replied, but they did not recommend that method. I would also want to know how many subjects, what proportion of your sample has missing data. Can you elaborate on that Tina? How many of your sample had missing data? Thanks
Deborah J Hilton, I have 22 variables in which individually 1 to 10% of datapoints are missing. Though there is one variable which has 41% missing data. In total around 6% of datapoints out of the whole dataset is missing.
As I said I can't recall Tina, the ? or answer whereby someone indicated not to use mean substitution, as I had always thought that was recommended. You could delete all the data with missing values, but then you introduce bias in that those who don't answer maybe a select group. Also the variable with 41% missing data, if you delete all of this, you remove a lot of your dataset, so that maybe problematic. I've just done a search on research gate using the term 'missing data' & there are a lot of very useful information answers, links and attachments. I would advise reading those ?/answers.
- Badges
- Science topic
- Similar topics
- Psychology