It seems to be human nature to want to use the most sophisticated tools on simple and complex projects alike. Perhaps it’s the feeling that using such complex solutions somehow produces “better” results. Or maybe there’s an element of “look what I can do” on the part of the user.
Whatever the cause, unneeded complexity for simple problems is almost always a situation where there is only a downside. If everything works, you’re no better off than the answer you would have obtained with a much simpler method. If anything goes wrong, you come away with a lot of egg on your face.
Certainly, many statistical procedures are very useful and can provide unique insight into relationships in the data. The wide availability of statistical analysis software has made the use of some of the most sophisticated techniques only a few clicks away. This easy availability has created a situation where the number of people who can implement a procedure far outnumbers the people who know how to implement a procedure, which leads to unclear, misleading, or just plain wrong results.
The solution? Keep it simple.
When looking at data and building an analysis plan, frame your thinking in terms of “what is the most simple way for me to see a relationship?”
For example, regression will show correlations between variables. Do you really need to perform a regression, or will a simple correlation table tell you what you need to know?
Cluster analysis will tell you how groups of records in the data are similar. Could you get what you need to know from a simple scatter plot?
Keeping things as simple as possible does a few things:
- Makes the results much easier to explain to others
- Reduces the chances of errors resulting from mis-application of the technique
- Eliminates the possibility of problems because conditions required for the technique to work were violated
- Reduces the amount of time needed to perform the analysis
The next time you are analyizing a dataset, force yourself to think in terms of applying the least complex methods to the data. The results may surprise you.