You’ve created a valid survey and gathered the responses, and now it’s time to analyze the results. Several advanced data analysis methods can help you and your statistical team uncover meaningful insights that may have otherwise gone unnoticed in your data.
Predictive modeling looks at how one set of variables predicts another variable. Two ways to do this are through driver analysis and targeting.
The method aims to determine the associated role different drivers play in a market. Let’s say one of your survey questions measures overall client satisfaction, and another measures satisfaction with different aspects of your company, such as quality, price and customer service.
You could use driver analysis to quantify the relative impact of each of the aspects of overall satisfaction.
Targeting aims to identify specific characteristics of a group that is of interest to your company. For instance, many companies want to know the demographic profile of their biggest customers.
Several different techniques can be used for targeting, although most of them are complicated and easy to misinterpret. Predictive Trees is the only method recommended for use by those without statistical expertise.
The goal of this method is to find groups of people who gave similar survey responses, with the overall aim of developing strategies that work specifically for that particular group of people. Let’s say one of your questions asked about hobbies. You could use segmentation to determine which group listed fly fishing and then create a strategy that worked to entice fly fishers to try your product or service.
Three types of segmentation are judgment, cluster analysis and latent class analysis.
Judgment involves using a single variable or small group of variables that relate to the key variables in your survey. If crosstab software revealed age was linked to many other survey questions, you could segment using age.
This traditional method of segmentation assumes:
- No data is missing, with respondents filling in all the variables
- All variables are numeric
- All variables have the same highest to lowest value, or the same range
The method often falls short because one or more of the assumptions is frequently not met.
Latent Class Analysis
This method is the updated version of cluster analysis. It uses advanced latent class analysis programs that can automatically overcome all the cluster analysis assumptions.
These charts illustrate the relationship between various categories in a survey. The map may be divided into four sections, with each section representing a different mindset. For example, the sections may represent characteristics such as:
- Openness to new things
- Air of rebellion
- Fondness for the traditional
- Calmness, innocence
Perceptual mapping would map out the survey responses in the four sections, giving you an overview of how customers perceived a particular product or service.
While there are various other data analysis methods you can use, these are some of the most effective and widespread ways to dig deeper and get more meaning from your results.