You are on page 1of 4

Market Research Expert TOP 10 Recommended statistical techniques for market research and customer insight 11May

There are a wide variety of statistical techniques that either form the main research process or act as a bolt on in order to glean maximum insight from a research study. Ci Research has over 20 years experience of applying advanced statistical techniques to survey data. We recognise that the language of stats can be daunting for some and, indeed, statistical analysis may not be relevant for a variety of research scenarios. We therefore only utilise advanced techniques when they will bring further valuable insight and will always deliver the output in layman terms. Most importantly, we also ensure that results are actionable with total clarity around outcomes and implications.

Our top 10 statistical techniques are as follows:

1. Factor Analysis Factor analysis is a useful data reduction technique, helping the researcher to see the woods for the trees. The technique groups issues together into a more manageable number of factors. Factor analysis can therefore help determine the underlying factors that summarises customer opinion from a whole host of product / service issues.

The technique is often used as a first stage in an analysis process. Factor analysis can be used to reduce the number of variables utilised to model customer behaviour, satisfaction etc. More concise variables will lead to a more robust model and an outcome, which can be easier for clients to interpret. Factor analysis can also assist the development of perceptual maps of customer opinion.

2. Regression Analysis Regression analysis is a widely used modelling technique and comes in a variety of forms. The basic aim of this technique is to establish causal relationships between a dependent variable, for example, sales, and one or more explanatory variables, for example consumer spending, season of year, interest rates etc.

The technique can be used to gain a snapshot of the drivers of a variable under analysis. A successful analysis will identify which issues best explain the movement in the dependent variable and so can help a client decide which parts of their business to concentrate upon in order to improve sales, satisfaction etc. Regression analysis is also commonly used as a forecasting tool.

3. Ridge Regression Ridge regression is a particular form of regression analysis and again seeks to identify explanatory variables that will produce an optimum model of the variable under examination. The technique is used instead of multiple regression when there are significant relationships between the explanatory variables, i.e. multicollinearity. This is particularly common in customer surveys, where certain issues are naturally related to one another. For example, a research survey may have five separate issues on the sales & ordering process. Respondents may be generally happy with the whole process and so give similar scores to the five issues. Multicollinearity would invalidate the inclusion of the five issues in a multiple regression model, whereas ridge regression may allow their inclusion if appropriate.

Factor analysis could be utilised as another way round this problem, although ridge regression allows the use of the original issue-level responses to give a more detailed analysis.

4. Logistic Regression Logistic regression is another form of regression analysis, this time to build predictive models for binary dependent variables, i.e. male/female, customer/non-customer etc. Although the model will again seek to identify the optimum explanatory variables, the main aim of the technique is to help predict membership of one group or the other.

By conducting an analysis on a customer sample, the findings of a logistic regression can help the client segment their customer database into prospect levels for example. The model can utilise all available customer information as potential explanatory variables and any binary category information e.g. infrequent / frequent buyer, as the dependent variable.

5. Discriminant analysis Like logistic regression, discriminant analysis looks to model a dependent variable that is categorical in nature, i.e. slow uptake, medium uptake, fast uptake etc. The analysis however may produce more than one equation or function that will assist in categorising respondents or customers into one group or another.

Logistic regression and discriminant analysis are, in some ways, interchangeable techniques. However, where logistic regression is more forgiving with different types of data used, discriminant analysis is more commonly used where the dependent variable has more than two categories.

6. Conjoint Analysis Conjoint analysis covers a variety of approaches that force the respondent to trade-off various attributes of a product and / or service offer. Two widely used conjoint analysis approaches are known as full concept conjoint analysis and pairwise analysis. The full concept

approach involves the respondent choosing between various full product/service offers either faceto-face or electronically. The pairwise approach can also be administered electronically, as the name suggests, comprises of the respondent choosing between different pairs of attributes.

The conjoint techniques are often used in new product development or more generally in determining the relative importance of various product / service attributes. As respondents are forced to choose between attributes / offers, more sensitive information about relative importance or worth is collected. Conjoint findings are particularly suited to aiding a successful cluster analysis.

7. Cluster Analysis Whilst factor analysis groups together different variables or research issues, then cluster analysis is used to group together different observations, i.e. respondents or customers. A cluster analysis will therefore look to divide an overall sample into sub-groups, where members of each particular sub-group share similar views or behaviour. Once clusters have been formed an analysis of the common characteristics of the members in each group can take place, allowing identification of behaviour amongst various customer types.

The technique is fairly adaptable in terms of the data it can use. As already mentioned, cluster analysis can be utilised to look at overall conjoint findings. Any meaningful quantitative data from a customer database or survey can however be used in developing clusters. Cluster analysis is commonly used for market or customer segmentation, so it is important to produce a valid solution whilst presenting the client with an actionable number of cluster groups.

8. Correspondence Analysis Correspondence analysis is a perceptual mapping technique that allows the plotting of both objects, e.g. retailers, suppliers, service operators etc. and attributes, e.g. service related issues in the same map. The analysis is similar to factor analysis in that it will seek to reduce the list of attributes into dimensions or factors that best capture the differing views of the respondents. Correspondence analysis takes this one stage further however by using these dimensions to plot objects and attributes in the same space.

It is important to note that the technique is descriptive and so further statistical techniques are required for inference or model building requirements. Correspondence analysis is useful, however, for helping to identify relationships that were not immediately evident and, indeed, is a widely used presentational tool.

The technique can be used to analyse non metric data commonly collected in research surveys such as Likert scales: strongly disagree through to strongly agree etc.

9. CHAID CHAID stands for Chi-Squared Automatic Interaction Detection and is used to identify the drivers of a particular dependent variable e.g. level of customer satisfaction. The technique splits the sample down into the key issues or factors that best determine action or opinion related to the dependent variable. A decision tree is thus constructed to summarise the issues on which the sample is split giving an easy-to-follow visual presentation. CHAID is useful either as a stand-alone technique or as an initial stage of a multi-layered statistical analysis to shed light on decision / action dynamics.

10. Structural Equation Modelling Structural Equation modelling (SEM) overcomes a fundamental limitation of previously mentioned multivariate analysis techniques that can only look at one set of relationships with a dependent variable at one time. SEM acts as an extension of multivariate techniques, in particular, multiple regression and factor analysis to allow the examination of a series of dependence relationships simultaneously. It is particularly prevalent when one dependent variable becomes a driver of another dependent variable. For example, what variables determine a companys image? How does that image combine with other variables to drive purchase decisions and customer satisfaction? And how does this satisfaction affect customers long-term loyalty?

You might also like