For the last few days we've been investigating ways to display the success of a marketing campaign divided into differently sized lists, each with a mailing and control group. In each constituent list, we have a mailing and control group, each with a size and a number of successes. We measure size as total sent mailings for the mailing group and total withheld mailings for the control group, and we measure successes as total sales conversions in each of the groups. For each mailing group, we also have figures for the number of respondents and the number of respondents who end up purchasing.
The challenge is to communicate all this information visually. After a few iterations, we've come up with the display below.
Each campaign is a two-step process. Ideally, the mailing leads subjects to respond out of interest. Then, it's up to the local store to wrap up the deal. The raw data.
A campaign's plot breaks down into regions, one for each list, delineated by rules over the list's label on the x-axis and separated by light gray bars. The overall lengths of each region are proportional to the total size, measured as total mailings plus total control subjects, of each list. Within each region are three bars. The first, colored light red, represents the proportion of respondents to the mailing--those who end up visiting a local store--as a fraction of total mailings. The second, colored a somewhat darker red, represents the proportion of conversions to sales from the mailing. There is also an overlap between the two, colored an even deeper red, whose width is chosen such that it's area, when compared to the area of either of the two bars, reflects the fraction of respondents who end up purchasing. The blue bar on the right side of each region reflects sales conversions among those in the list's control group. The relative size of the mailing group versus the control group is represented by the relative widths of the bars. It's worth noting that each bar has an arrow above and beneath its peak; those represent 90% confidence intervals about each proportion, computed by binom.test.
It was a real challenge to develop a display for this type of marketing data. If we have a campaign with many lists, each a typical case-control construct, how do we represent it visually? We've looked into funnel plots like those drawn by rmeta, but they just don't have the impact we want, not to mention that they ignore all sorts of useful information, like the relative sizes of mailing and control groups, relative proportions of successes (more intuitive than the odds ratio) and standard errors. If anyone has any bright ideas or suggestions on how we might improve this display, please let us know by commenting on this post (the "Comment" link). We'd love to hear from you!
Recent Comments