What is happening to those hapless survey takers?

Life insurance actuaries, the Census Bureau, and the Social Security Administration use a tool to understand mortality: life tables. Life tables measure the odds of dying at a particular age. Age 0 to 1 has a higher risk of death than age 1 to 5. Past age 5, our odds of death steadily climb.

One might assume that a similar pattern would be true for online surveys, and that the odds of abandoning a survey increase with each increasing page. So, I decided to check it out. Do survey abandonment curves match similar patterns to human mortality tables? I used one of Allegiance’s client surveys to examine the trends.

In the chart below I used one axis to plot the number of survey takers who had completed each page. On the secondary axis, the red line, I plotted the odds of abandonment. The odds of abandonment is the number of survey takers lost to page x+1 divided by the survey takers remaining at page x.

Turns out, survey abandonment curves are not quite as smooth as human mortality curves. Survey abandonment curves are spiky. It’s like the Grim Reaper inhabits certain pages but not others.

Clearly, one can see where the problem pages are, where abandonment spikes. It turns out that this survey has a bunch of open-ended questions on pages 2, 3 and 10. Answering open-ended questions is a lot of work. Rather than work, people abandon the survey altogether.

There is no way this client will get rid of their open-ended questions. There is nothing I can do to make the questions less prone to abandonment. However, what we can do is re-order the pages and put the low-abandonment pages at the front of the survey and the high-abandonment pages at the back. The thought is that with more survey takers sticking around for longer, we may get to gather more data.

We can measure the increased data the same way the census bureau measures “life-years.” I’ll call this measurement “page-completes.” If we accumulate more page-completes by the end of the survey then we’ve made an improvement. Below is a simulation of re-ordering the pages to put the high-abandonment pages at the end.

Sure, the same number of respondents made it to the end of the survey, 1910. However by re-ordering the survey, the page-completes grow from 28,828 before to 32,319 after. Re-ordering may help me capture more survey data overall.

Kyle LaMalfa, Best Practices Manager and Loyalty Expert, Allegiance