Tips, Trips and Troubleshooting for Maps
Using Q to analyze other data tables
Q can conduct correspondence analysis on a non-Q table by saving the table in Excel as a CSV file, reading this into Q, creating a Number – Multi question of the table (excluding the row names), selecting the table in the blue drop-down and RAW DATA in the brown drop-down and creating the map. The output will show row numbers which then need to be replaced with the row names.
Analyzing numeric and other forms of data
Although the theory of correspondence analysis is designed with Pick One questions (i.e., categorical variables) in mind, it can be used to analyze any data tables. There are some general principles to keep in mind:
- It is generally advisable to have data tables that show things in the same scale. For example, one column showing percentages and another showing millions of people will likely be problematic.
- Often the most valid approach is to quantize variables (e.g., turn a Number question into a Pick One question and collapse categories).
You can create the chart either from RAW DATA or a table.
Overlaying additional information onto maps
When using multiple correspondence analysis, you can select Save factors to created additional variables. These additional variables are scaled by the canonical correlates (i.e., they are principal coordinates). You can add additional information on the map by crosstabbing these factors with other variables. The resulting means will show the appropriate coordinates on the map.
Using multiple correspondence analysis to form segments
There are two basic ways to use correspondence analysis to form segments:
- Using judgment. For example, with a map showing brand associations, judgment can be used to group together brands that are on similar positions on the map, and consumers can be assigned to segments based on their relationship with the brands (e.g., consumers can be grouped into segments based on the brands that they buy or like).
- When using multiple correspondence analysis, the factors can be saved and then segments can be formed using either judgment, cluster analysis or latent class analysis (This is because approaches such as this introduce two separate forms of error into the modeling – the error in forming the factors and the error in forming the segments – whereas latent class analysis only has the one form of error and will thus have less error in the model; furthermore, the mathematics of latent class analysis guarantee it will usually not have more error than cluster analysis). Where the intent is to use judgment, this approach can be appropriate, but in general it is preferable to use latent class. When using the factors from multiple correspondence analysis, it is important to remember that each factor has a smaller standard deviation; the easiest way is to form segments using latent class analysis and set the Distribution to Multivariate Normal – Diagonal.