Hello, I am doing data analysis with PLABSTAT, after the analysis I am getting information about the suspected outliers. So, how can I deal with it? Should I delete those values and regard the observations as missing values? or is there any other way to deal with those?
What makes you think it is an outlier? SUppose you have 100 data points, There should be 0-2 data points that are 3 standard deviations from the mean. If you have skewed data, you may be fitting the wrong distribution to the data.
If you delete "outliers" because they are outliers, you do yourself and your analysis an injustice. If your data comes from a medical study, you could be doing something criminal.
Hi Anik
One needs to think this through a little more in light of the complexity of the data and assumptions. Try this website to start you with the various options to help you to crystalise the process.
http://www.theanalysisfactor.com/outliers-to-drop-or-not-to-drop/
Cheers
Roman
What makes you think it is an outlier? SUppose you have 100 data points, There should be 0-2 data points that are 3 standard deviations from the mean. If you have skewed data, you may be fitting the wrong distribution to the data.
If you delete "outliers" because they are outliers, you do yourself and your analysis an injustice. If your data comes from a medical study, you could be doing something criminal.
The answer might be specific to the analysis being performed.
If outliers are at the same time very influential on the estimates then you might think of finding a way to handle them. There are many options like using another robust method to outliers, keeping them etc. Deleting them should be the last option.
Before any data analysis you must do Quality Control of data to be used. This includes detecting outliers. I am not sure how PLABSTAT works and the type of data you are analysing. You might have to look at a guide on how to handle the outliers in this package. Here you could even contact the person or organisation who gave you the data. Note that deleting outliers is your z option for your analysis.
I suggest that you consider methods that are robust to outliers. See, for example, Maronna et al, Robust Statistics, Wiley. The idea is to identify the outliers and to try to understand how they relate to the rest of the analysis. As Mehmet says, simply discarding them is usually a bad idea because they often represent useful information. Best wishes. BTW, it is easy to run a robust regression analysis. See Venables and Ripley, Modern Applied Statistics with S, 4th edition, Springer.
Dear Anik,
1st you use box plot diagram for identifying the number of outliers. This box plot diagram contain box and outlier information, but not all outliers are effective. You should remove only effective outliers.
See this paper:
Hybrid functional link artificial neural network approach for predicting maintainability of
object-oriented software
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis