|
A fairly common, though often misunderstood practice in marketing research is the weighting of survey data. To the uninitiated, weighting can seem like statistical sleight of hand intended to misrepresent data or to support predetermined conclusions. To those who utilize this important tool, however, weighting is an essential component to providing accurate, representative data. So what is weighting, why do we do it and how is it done?
Weighting Defined
Simply defined, weighting is the attempt to alter data to reflect truer population proportions than were encountered in the data collection process. For example, suppose that the entire world consists of Blue people and Green people. Suppose further that we have completed surveys with one hundred Blues and one hundred Greens. Our findings: 80% of the Blues like donuts compared to just 40% of the Greens.
It doesn't take a statistical genius to sense that there may be a significant difference here. However, while we can use this data to determine differences between these groups, we can't just look at the survey totals of both groups and say that 60% of the world (the average of the two percentages) likes donuts. What if Blues represent 90% of the population and Greens only 10%? Ensuring that this "total" measure properly portrays the real world is what weighting is all about.
Why Do it?
You may be asking yourself, "Why not avoid this whole weighting thing, implement quotas from the beginning and complete 90% of your interviews with Blues and 10% with Greens? Or simply employ a random sample design?"
While this would work just fine to read the data in total, if your goal is to also analyze differences between groups, you probably won't have enough completed interviews with Greens to contrast their responses with Blues. Two hundred interviews completed randomly or with quotas for Blues and Greens would yield approximately 180 and 20 completes for these groups, respectively. But you really can't expect to perform any meaningful analysis on just 20 Greens. "Increase the total number of interviews and go random," you say? Remember that you would have to complete 900 interviews with Blues to get a more friendly, analyzable number of 100 Greens, and this would have an adverse effect on your budget, to say the least.
How To Do It
Now, properly motivated, you ask, "So how is it done?" First, having hopefully identified the true proportion of Blues versus Greens in the overall population, we divide that number by the survey proportion and we have our weighted value.
| Group |
True
Proportion |
Survey
Proportion |
Equation |
Weight |
|
| Blues |
90% |
50% |
90/50 |
1.8 |
| Greens |
10% |
50% |
10/500 |
.2 |
At this point tabulation software can take over, and the running of crosstabs will generate numbers that can be analyzed with confidence that they represent the "real world" by allowing the responses of Blues to "count" more than Greens by exactly the weights determined. Oh, so back to answering our original question, the "weighted" data reveal that 76% of the Blue and Green world likes donuts.
|
