Statistical techniques in market research
An explainer
Within the market research industry there are a number of statistical techniques we can use within quantitative research to better understand and get deeper insight our respondents. However, these techniques and tools are not easy to find out about, information is hidden and not well explained. This page will help you understand some ways we can make research better.
Two issues explain why we need to look deeper into our data and apply statistical techniques:
1. Respondents are not one dimensional clichés - people aren't defined by their demographics.
2. Respondents are not machines, they don't answer our questions in a logical manner, they don't consider all responses or questions equally; and they can't articulate what is really driving their choices.
The techniques below show some of the ways we can help mitigate these two problems:
The techniques
Key driver analysis
Key driver analysis uses regression analysis (mini-machine learning) to find which factors influence an outcome. This technique goes beyond looking solely at correlation of variables but looks at the interaction between all input variables and cause of the outcome. Which input variable drives the outcome - it tells us what is really important to a respondent not just what they say is important.
Output: Derived importance scores for all input variables
Use: Allows the client to prioritise factors for business and marketing strategy
Bayesian Networks
Bayesian networks goes many stages further than key driver analysis and looks at what drives the key drivers and can go back many steps on the buying journey. It gives us a greater understanding of the factors influencing the outcome, the pathway to purchase.
Output: A network map showing how input factors lead to a desired outcome.
Use: Allows the client to prioritise factors for business and marketing strategy. Gives them more of an understanding of their customers pathway to purchase and allows more opportunity to influence this outcome.
TURF analysis
TURF (Total Unique Reach & Frequency) analysis is a different way to interpret brand and product data. Rather than looking at preference share we view which brands or products get us additional customers and which are being bought by those already choosing other products - which products extend our reach. Which new ice cream flavour will bring new customers in?
Output: Chart showing unique preference share
Use: Allows the client to identify which additional products or range derivatives will extend their customer base and which will result in no new customers.
Cluster segmentation
We can segment customers in 4 ways; Geographic, Demographic, Behavoural or Attitudinal - or a mix of these. Using cluster segmentation allows us to build segmentations based on the responses to our survey - how respondents group naturally based on their attitudes and behaviour. This allows us to build groups based on who our customers really are rather than using stereotypes and cliched demographics. We cluster based on k-means or latent class analysis - techniques that use machine learning, means and distance to find the best fit for grouped data. These clusters can then be applied to the survey and explained exploring their similarities and differences.
Output: Segmentation report - defining characteristics, attitudes; market share, product preference. Pen portraits.
Use: Allows the client to better understand their market, align and explain their business and marketing strategies - get internal buy in.
MaxDiff
Humans are not very good at ranking or choosing from long lists but find it easy to choose their favourite when presented with one or two options. MaxDiff is a technique that reduces a long list of options to a range of choice statements - the respondent is presented with 2 or 3 options and chooses their favourite, this is repeated many times; the data is then aggregated and analysed to correctly rank our inputs.
Output: Ranking score for options (brands, products, statements etc)
Use: The client can accurately understand where each option ranks relative to it's competitors
Conjoint
Another difficulty we face is that respondents are NOT able to accurately identify or explain how important constituent parts of a product or service are. To help us get passed this and identify how and why customers choose a product we can use conjoint analysis. In conjoint analysis we separate the product into it's constituent attributes, these attributes are split into levels. Then, as in MaxDiff, we show the respondent multiple sets of products derived from the attributes and levels. Analysing the choices of these derived sets means we can put a value on each attribute and each level - how important are they in driving customer choice? This is the best method understanding pricing.
Output: Ranking score for attributes and levels, interactive market preference share dashboard
Use: Understand which attributes most impact customer choice - product/range development. Understand the impact of price changes.
Van Westendorp pricing model
Van Westendorp modelling is a simple method to better understand and identify a product's optimum price point and an acceptable price range. They are derived by asking 4 questions:
1. What price it too cheap?
2. What price is good value?
3. What price is expensive?
What price is too expensive?
When presenting the questions to the respondents it is important that they are educated in the market, they know what a product should roughly cost; because of this we may need to prompt or provide competitor costs. This better matches the buying process and leads to better results.
Output: Optimum price and acceptable range across customer segments and markets
Use: The client understands their product pricing position and it's elasticity.
Pricing - Gabor Granger
Another method of determining optimum pricing and pricing curves. We ask the respondent whether they would purchase a product at a certain price, if they say Yes, we ask again at a higher point; if they say No, we ask again at a lower point. We continue until we reach the point the respondent will purchase. With this data we know at each price point what % of the population will purchase and so total profit at each price point.
Output: Optimum pricing for purchase % and profit
Use: Allows the client to maximise sales and profit
Want to know more?
We offer bespoke training sessions to research and data teams explaining and training on the optimum way to use the tools above.
For more information on this, to book in some training or to speak about the statistical support services we offer get in touch using the link below: