Main Content Area
Stop Using Lower Numbers to Mean More of Something!
Why is it that some researchers code their scales such that lower scores mean more of something and high scores mean less of it? I have seen this in the research industry as well as academia. For example, this would happen if I was measuring brand attitude and the verbal anchors for the response scale were scored such that lower scores meant a better attitude. This can occur at two critical points in the measurement process: 1) when respondents answer the question and are explicitly shown that a 1 is used to represent something high such as better attitude, greater agreement, more usage, etc.; and 2) when the analysis is done and reports are written.
With respondents, higher numbers usually imply more of something, so it can be confusing if the question has the numbers next to alternative answers and lower numbers are paired with answers that actually mean more of something. That can be eliminated, however, with online surveys by just showing the verbal anchors of the scale rather than numbering them. In other words, instead of telling respondents that in the next 10 questions they are to check 1 if they strongly agree, 2 if they agree, 3 if they are neutral, 4 if they disagree, and 5 if they strongly disagree (which would be confusing), just design it so there are radio buttons next to the verbal anchors, no numbers.
That still leaves the problem of what happens during the analysis. If lower numbers are used to mean more of something then that can lead to incorrect interpretation of results by those reading reports who are not clearly understanding the coding. For example, if I report that men scored 2.5 on a brand attitude scale and women scored 4.3 then readers would naturally assume that women had better attitudes, but they would be entirely wrong! There is also the potential for statistical problems when some variables use higher numbers to mean more of something while other variables use lower numbers to mean more of something. For example, let's say you wanted to measure the relationship between age and brand attitude. Let's also assume you have respondents' actual ages; thus, higher numbers mean greater age. When you correlate age with your brand attitude scale (that uses lower numbers to mean better attitude) and find there is a significant negative correlation it is easy to report (if you are not being very careful) that there is a negative relationship between age and brand attitude. But, that is exactly the opposite of what is true! While there is, indeed, a negative correlation between your measures, it is because of the backwards way brand attitude was coded rather than there actually being a negative relationship between the constructs themselves.
So, the bottom line is that whenever possible, code answers for questions and scale items so that higher numbers mean more of the construct. You will be doing respondents and those who read your results a favor.