We spent years pioneering and refining our proprietary Unbounded Scale. It allows us to capture desirability in a very powerful way.
Closed-ended scales require respondents to circle a number or check a box. Scales can be anchored with descriptions such as "not important", "very important", "strongly disagree", "strongly agree", and so on.
Early on, we observed significant problems associated with bounded scales.
One problem is that, because there are end-points, once someone has given the "highest" rating to one attribute, nothing else can be measured as being more important to them. This problem is dealt with in traditional research by rotating questions. However, functionally this "solution" serves only to randomize the error, it does not lessen the problem. With a reasonable number of questions, we can be relatively sure that for the vast majority of respondents closed-ended scales fail to identify that which is most important to them.
Another problem is that closed-ended scales show "bunching" at various points. This bunching effect is an artifact of the scale, not the information we are trying to measure. It introduces error.
Keeping in mind that the measurements we take serve as model inputs, and that better inputs yield better output, examine the graphs below. They demonstrate the power of our proprietary Unbounded Scale.
Scores collected using EMA’s Unbounded Scale show normal distribution within and across attributes.
Desirability using EMA's unbounded scale
Scores collected using closed-ended scales show "bunching" and are not normally distributed, indicating an improper measurement is taking place.
Closed-ended scales require respondents to circle a number or check a box. Scales can be anchored with descriptions such as "not important", "very important", "strongly disagree", "strongly agree", and so on.
Early on, we observed significant problems associated with bounded scales.
One problem is that, because there are end-points, once someone has given the "highest" rating to one attribute, nothing else can be measured as being more important to them. This problem is dealt with in traditional research by rotating questions. However, functionally this "solution" serves only to randomize the error, it does not lessen the problem. With a reasonable number of questions, we can be relatively sure that for the vast majority of respondents closed-ended scales fail to identify that which is most important to them.
Another problem is that closed-ended scales show "bunching" at various points. This bunching effect is an artifact of the scale, not the information we are trying to measure. It introduces error.
Keeping in mind that the measurements we take serve as model inputs, and that better inputs yield better output, examine the graphs below. They demonstrate the power of our proprietary Unbounded Scale.