Last week’s tutorial covered the basics in online survey design, and this week’s tutorial covers the basics about analyzing survey results. This handy decision tree with notes shows which tests should be used for analyzing various data sets.
Continuous data comes from the opinion questions on your survey, which have scales that participants use to indicate their feelings about the subject such as:
Please use a scale of 1 to 7 to rank your preference for strawberries.
Categorical data help identify demographic information about survey subjects:
Do you grow strawberries?
Most datasets are imperfect. There are usually cases of incomplete information from users who began taking the survey then leave, there will probably be outliers and “flat-liners” who use a single number to answer all continuous data. Using your own best judgement (on very large datasets, cleaning may not as necessary), you will want to clean the dataset. Box plots are very helpful in identifying cases of extreme outliers for each question. Reliability tests should be performed on each construct, and unreliable questions (Cronbach’s Alpha of 69% or less) should be dropped.
Each test reveals relationships between constructs. The predictive ability of the survey lies in the strength of these relationships (A website that uses predictive survey relationships in very exciting ways is www.hunch.com).
Ultimately, predictive models developed from test outcomes can be a very useful marketing tool crafted from survey results, provided that they are accurate more than 50% of the time.
Most of these models find their way into CRM policies. If a company can predict that a current customer is more likely to switch providers based on the relationship between their method of acquisition and perception of quality, then when communicating with that customer a customer service representative will focus on quality assurance. Other uses include the amount of money a company will invest in retaining desired demographics of customers, and identifying “problem customers.”
Of course, there are simpler ways to interpret survey data, but this is a good basic framework for exploring the statistical applications of survey data.
Was this article helpful? Have you used statistical analytics of survey data in particularly exciting ways? Please share your comments about survey data analytics with the community at Marketfix.