Scales - Think Visually

We've seen it in telephone interviews since the beginning of time... Survey designers attaching numbers to verbal scale questions. This is often done because numbers are easier for respondents to remember than words are.

For example, let's say you want respondents to rate 5 statements using a scale that consists of:

• Strongly agree
• Somewhat agree
• Neither agree nor disagree
• Somewhat disagree
• Strongly disagree

If you've ever monitored a telephone survey, you realize that some respondents simply have a hard time remembering the choices. You'll get answers like "I agree", "I think that's true", or simply "yes". As we all know, those don't answer the questions in a way that are comparable to others answering the questions, so interviewers are forced to ask respondents for clarification... Often over and over and over.

In an attempt to combat this problem, some survey designers will assign numbers to each response choice. An example of this type of question wording might look something like this:

"Next I'm going to read you a series of statements others have made. For each, I'd like to know how much you agree or disagree. Please use a scale where 1 means strongly agree, 2 means somewhat agree, 3 means neither agree nor disagree, 4 means somewhat disagree, and 5 means strongly disagree.

The first statement is...

...Wow... What a mess! I don't know if this helps or hurts (I suspect it doesn't help), but I do know there is a better way to do it on the Web.

When designing a Web survey, you have to change your thinking a little bit. You have to remember that the Web is a visual medium. You do not have to worry about whether or not the respondents will remember the words because you can display them on the page in front of them. So, for Web display, the same question might look like this:

"Below is a list of statements others have made. For each, please rate how much you agree or disagree."

This is a much cleaner solution to those long question scales, and will yield more consistent results between respondents.