‹ How to analyze survey results
What is ordinal data
Among all kinds of datatypes, ordinals are very popular. Almost all questionnaires include some questions using ordinal type. As an example, when we deal with the level of supporting/opposing an opinion/fact/statement, the responses usually have the form: completely agree – agree – somewhat agree – disagree – completely disagree.
As it is clear this type of data is very popular and probably even your questionnaire has already included many questions with ordinal type answers. However you need to be careful here, the analysis of ordinal data could be tricky.
There are different methods to summarize ordinal data. As mentioned before, we usually require to convert ordinal data in such a way that can be handled using the underlying statistical package.
One way to do this is to implement a numerical scaling technique (e.g. 5-point scale, 7-point scale, etc.) which devotes a unique number to each ordinal e.g. completely agree (5) and completely disagree (1). Hence you will be able to refer and use this ordered numbers. For example, you can calculate an average value for each questions and use it to analyze the results.
In case you use N/A as one of the answers, you can assign number 0 to it, in your statistical calculations.
However, sometimes using this scaling methods could be really misleading. Suppose you use a simple arithmetic mean (average) on your scaled data, and considering it as a representation of your data. In this case, your analysis probably suffers from a misjudgment. In order to clarify it consider a case that we have a 50%-50% extreme responses (completely agree (5), completely disagree (1)). The average is about 3 which represent somewhat agree. Nevertheless, you already know that there is a big gap between your population’s opinions, and therefore, the average (somewhat agree) can not represent the reality of your population.
In order to avoid these type of mistakes, some experts suggests to use contingency tables (like nominal data) to show each result. This method would be helpful for descriptive statistical analysis and it help to avoid of using such a misleading inference that can cause by using arithmetic mean. Additionally, you may use diverging bar chart to visualize your ordinal data. This method gives you a good opportunity to summarize your descriptive analysis visually.
However, if you would like to go deeper in your data and compare the relationship among your ordinal data statistically, you can apply some more sophisticated methods like Spearman’s rank correlation.
All in all, it is important to adopt a proper method to convert and handle ordinal data and most of the time it could be really misleading to summarize your data by considering a simple arithmetic mean.
More about other data types
Nominal data Numerical (ratio) data Interval data