Welcome to my first post in this six-part series. Over the series, I’ll explain methods of dealing with survey data and ways to visualise this in Tableau. In Part I, I’ll run through some of the core theory surrounding survey data.
The Series
- Types of survey data inputs (including Likert)
- What does the mean mean? (coming soon…)
- Prepping survey response data (to-do)
- Creating Likert scales in Tableau (to-do)
- How to represent Likert on a graph (to-do)
- How to make the graph (to-do)
Background
We had a client project week recently at the Data School which involved survey data. Survey data comes up in every industry that involves people, which means it comes up in every industry. However, for non-formally educated statisticians like myself, it can be a lot to take in. So I wanted to break down some of the key insights I learned over the project week for this six-part series.
The Basics
We can break down surveys into two parts;
- Demographic questions
- Survey questions
Demographic questions let us cluster the data and enable us to make meaningful insights. We can also use it to check whether we have a fair sample and aren’t underrepresenting a particular group. For example, “What is your age?”.
Survey questions relate to what we’re actually interested in finding out. If the survey was interested in streaming entertainment, you might see a question asking “How many hours do you spend watching Netflix?”.
The Three Main Types of Questions
This brings us to the next point – there are three main types of questions in a survey:
- Single Response
- Multiple Selection
- Likert Scales
A Single Response question is a simple yes or no response. In keeping with our streaming theme, a question might be “Do you own a smart TV?”. Multiple Selection is fairly self-explanatory – you get to choose from a list of responses.
I’ll discuss Likert further on. The reason these are the main types of survey questions is that they can be easily aggregated. Aggregation is necessary because it means the data is easily anonymised and also allows us to create data visualisations other than word clouds.
Numerical questions where the expected range is quite narrow (such as “How many cars do you own?”) are not uncommon, but even these are more likely to be asked as Multiple Selections (often with “5 or more” as one of the responses). Knowing that 2% of the population have more than five cars is a more valuable insight than knowing 0.0002% of the population have 11 cars. The caveat is that this can also go the other way – knowing that 60% have at least one car is less valuable than also knowing 40% have one, 5% two, 10% three and so on.
You can always aggregate up; you can’t aggregate down.
Likert Scales
In case you don’t already know what a Likert Scale is, it’s this thing:
- Strongly disagree
- Disagree
- Neither agree nor disagree
- Agree
- Strongly agree
A Likert Scale is used to get a respondents sentiment. Invented in 1932, it’s well established and still used often in surveys. In my next post in the series, I’ll discuss the Likert Scale a bit more with an emphasis on how it should be interpreted.
Often in surveys, Likert styled questions are reworded and repeated to give a highly correlated set of questions. For instance, an answer to “I prefer dogs over cats” should inversely correlate with “I prefer cats over dogs.”. A cat person is likely to disagree with the first and agree with the latter. Asking multiple questions which correlate allows the data to be aggregated to give an understanding of the sentiment of the overarching topic. If interested in learning more on this, the Wikipedia article (https://en.wikipedia.org/wiki/Likert_scale#Composition) has a well-written section on Likert composition.
Summary
Surveys have both demographic and survey questions. Questions are commonly styled as Single Response, Multiple Selection or as Likert Scales. The reason we restrict the responses with these questions is to make aggregating the data more straightforward. However, in this process, information is easily lost, and care needs to be taken as aggregation is a one-way ticket.
In the next part of the series, I’ll go into a bit more depth and cover topics such as what the mean means in Likert scales and the significance of nulls in survey data.
As always, thanks for taking the time to read my blog! If you have any comments, suggestions or want to chat, free to connect with me on my LinkedIn!
~ Ryan Edwards