Our analytical services add insight through multivariate analysis and other techniques across a range of methods. We regularly use these approaches to dig deep in to survey data, adding insight and value.
- Regression analysis
- Key Driver Analysis through appropriate methods (regression, MaxDiff, random forests).
- Conjoint analysis
- Maximum Difference Scaling
- Tree and Forest models
- Cluster analysis
- Segmentation of groups through appropriate methods (cluster analyses, latent class analysis)
- Random Forests
- Log-linear modelling
- IRT models
- Correspondence and multiple correspondence analysis.
- Missing value imputation
- Analysis of Variance
MaxDiff Case Study
Over recent months we have helped a social housing provider come up with new home designs using Maximum Difference (MaxDiff) Scaling.
MaxDiff is a technique for measuring the importance or preference of multiple items. In MaxDiff tasks, respondents see sets of items (typically 4 to 6). In each set, respondents indicate which item is most important (preferred) and least important (preferred).
MaxDiff is shown to provide results that have greater between-item and between-respondent discrimination, and greater predictive accuracy than either monadic ratings or paired comparisons.
Asking respondents to rank large numbers of items is unreliable, as is asking a scale question for each item.
Ranking tasks become difficult to manage when there are more than about seven items, and the resulting data is on an ordinal scale only.
Rating tasks assume that respondents can communicate their true affinity for an item using a numeric rating scale. Rating data is often negatively affected by lack of discrimination among items and scale use bias (the tendency for respondents to use the scale in different ways, such as mainly using the top or bottom of the scale, or tending to use more or fewer available scale points).
MaxDiff questionnaires are relatively easy for most respondents to understand. Furthermore, humans are much better at judging items at extremes than in discriminating among items of middling importance or preference. Also since the responses involve choices of items rather than expressing strength of preference, there is no opportunity for scale use bias.
After analysis, MaxDiff produces a rank between the items tested plus a metric distance between the items.
The output for a MaxDiff exercise is an interactive Excel tool. This can be customised to suit the clients’ requirements. Filters can be added to the tool so that comparisons may be made across different sub-samples.
Key Driver Analysis case study
To supplement the analysis of its 2013 employee survey data, our health sector client commissioned the application of Key Driver Analysis (KDA). KDA seeks to determine the key influences on a number of crucial variables within survey data.
Multiple Regression is the statistical technique which underpins the Key Driver Analysis (KDA). The analysis is based on the comprehensive range of attitude scales contained within the survey questionnaire. The objective was to find the relative impact of individual aspects of attitude on key indicators including overall satisfaction with the job.
The principle of the method used in KDA is based on assessing the statistical correlation between employees’ ratings on each of the possible “key drivers”, in turn, and their ratings of overall satisfaction with their job (plus how well change is managed, feeling valued for the work done and feeling valued and recognised for contributions made). This correlation is based on the proportion of variation in overall satisfaction that could statistically be “accounted for” or “explained” by related variation in satisfaction with individual factors. If the correlation is high, then the factor will be “important” in the sense of the analysis. If it is low, it will imply that the factor is less important. The rationale for this is that a high level of correlation implies a likelihood that improving satisfaction levels for the individual factor will in turn improve overall employee satisfaction. If there is little or no correlation, this offers no evidence that improving the factor might have any impact on overall satisfaction. We reflect the importance as an index value in which 1.0 is equal to the average importance across all factors. A level of 2.0 implies that the factor in question is twice as important as the average. “Key Drivers” are factors that have high importance.