This list comes from the Coursera Data Analysis Course.
Linear and Logistic Regression are some of the most common techniques applied in data analysis. Here is a list of possible problems with regression in the real world.
- Confounders – variable that is correlated with both the outcome and other variables in the model
- Complicated Interactions – how do the covariates interact
- Skewness – is the data not evenly distributed, heavy to one side or the other
- Outliers – data points that don’t fit the pattern
- Non-linear Patterns – not all datasets can be fit with a straight line
- Variance Changes
- Units/Scale issues – make sure the units are standard across the model
- Overloading Regression – too much complexity
- Correlation does not imply Causation
What other problems do you find when using Regression on real-world data
Do you know of other problems that are missing.
Small Sample – absence of sufficient data to fit a regression model.
That is a common problem. Professor Jeff Leak did not add that to his list. I wonder if that problem is not specific to Regression, because all statistical/machine learning models suffer when not enough data is present. I would agree with you though; small sample size can be a problem when doing any data analysis.
Thanks for commenting,
Ryan
[…] http://101.datascience.community/2013/03/07/9-problems-with-real-world-regression/ […]
[…] http://101.datascience.community/2013/03/07/9-problems-with-real-world-regression/ […]