Research projects can become complicated in a hurry. There is a lot to keep track of: complex business problems, data quality issues, different modeling techniques, multiple data sources, and so on. To keep your project on track, it helps to have an well defined methodology in place.
There are many different types of projects you can do in Displayr and each has their own unique problems. For example,
- Driver Analysis
- Text Analysis
- Market Segmentation
- Correspondence Analysis
- etc.
The wrong approach is to simply read your data set into Displayr and start to haphazardly run procedures. It is very easy to run complicaed procedures in Displayr, but unless you have a strategy, you will just end up wasting a lot of time.
Most successful research projects involve following a set methodology. For example:
- Define your research objective
- Prepare your data for analysis
- Select the appropriate analysis technique
- Choosing your missing value strategy
- Reviewing diagnostics
- Visualizing the results
The sequence of these steps is not strict. Many times, you will need to move back and forth between steps as necessary.
In this article, we will summarize the steps performing a Driver Analysis. While some of these steps are unique to a Driver Analysis, all of these steps can be adapted for any type of research project.
Requirements
A research project you would like to peform
Method
- Define your research objective. For example, you want to know the relative importance of the different drivers that determine satisfaction in your company
- Prepare your data for driver analysis.
- Make your predictors numeric of binary. The standard driver analysis techniques assume that the predictor variables are numeric rather than categorical variables.
- Assign higher values to better performance levels of the outcome and predictor variables, The standard driver analysis techniques assume that the outcome and predictor variables are ordered from lowest to highest, where higher levels indicate more positive attitudes.
-
Stack or use auto-stacking software (if you have repeated measurements). Driver analysis assumes that there is one variable (i.e., column) containing the outcome variable and one variable for each predictor variable.
- Make your predictors numeric of binary. The standard driver analysis techniques assume that the predictor variables are numeric rather than categorical variables.
- Select the appropriate analysis technique. To do a driver analysis, you need to select the right regression technique. Many potential outcome variables can be used when conducting a driver analysis, from five-point rating scales to utilities from conjoint studies. To perform a valid driver analysis, we need to select an appropriate regression type for our data. If you select Linear, we recommend Shapley Regression or Johnson's Relative Weights.
- Select your missing data strategy. Before deciding on how to deal with missing values, it is a good idea to analyze the extent of them and whether or not there is a pattern. For example, it may be that one or two variables may have a huge percentage of missing values, in which case, it is usually a good idea to drop those variables from your model.
- Review diagnostics. Techniques like driver analysis make many theoretical assumptions. It can be useful to verify that these assumptions are appropriate. Displayr and Q automatically perform common checks. Where the checks do not pass, they are shown in orange boxes.
- Visualize the Results There are typically three main types of data visualizations used in driver analysis:
· Visualizations of importance scores· Importance by sub-group· Quad-maps / performance-importance scatterplots
Next
How to Get Started with the Displayr API