4 Out of 5 Marketers Recommend This: Lessons on Statistical Interpretations
Colgate ran a classic ad in the early 2000s that received a lot of attention: 80 percent of dentists recommended Colgate toothpaste. Sounds like a stellar endorsement, right? Four out of five dentists would prefer you use Colgate over any other toothpaste — surely Colgate must be the best. But the attention the campaign drew was not positive. In fact, a statistical scandal came to light and put Colgate under fire.
The company surveyed dentists to come to this statistical conclusion, but had manipulated the survey and resulting statistics to work in its favor. Dentists in the survey were actually allowed to select several toothpastes simultaneously when asked “Which toothpaste would you recommend?” Thus, the 80 percent statistic was not mutually exclusive as the wording in the ad suggested.
Statistics are finicky: If you are not attuned to their subtleties and trickeries, you may be none the wiser. But the more aware you are of statistical interpretation, the lower your risk of implementing faulty stats in your own marketing.
One Dimensional Numbers
A statistic is virtually useless on its own. Context is key — the more you can gather, the clearer picture you will have. Let’s say you are running a campaign that is targeting 2016 voters, and are looking into previous data to get an overview of your targets. You come across this statement:
“An estimated 57.5 percent of eligible voters turned out to vote in 2012.”
What does this statement tell you? What have you learned? What can you infer?
Not much of anything, really.
Relativity is the best descriptor: With no information to give context to this statistic, there is no way you could apply the information to help your campaign. You need some external variables to get any sort of idea of what this number really means for you.
Is it war or peacetime? Is the current president a candidate? What was voter turnout like in the election before 2012, or the ones before that? Have any voter registration requirements changed? How different are the candidates from each other? What were the demographics of voters versus the whole population? What has the political climate for 2016 been like? What is different between now and 2012?
These types of contextual questions will make your statistics and analyses more reliable — they will yield a more accurate prediction of expected voter turnout and demographics for 2016. Now you have a better idea of which regions will result in higher ROI from advertisement expenditure and which segments you should most heavily target.
Just like you want a 360-degree view of your customers, you want a 360-degree view of marketing building blocks.
Wily, Willful Words
Consider this statement from a survey report:
“Additionally, even consumers don’t want to deal with paper coupons: 76 percent say digital coupons are a convenient option compared to printed coupons, and 47 percent of shoppers say they wish all coupons were digital.”
If I read either of these two sentences individually, I would interpret them to have the same implication: The percentage reflects how heavily I should focus on digitizing my company’s coupons. But juxtaposed, you see that there is a substantial difference in both the wording and the corresponding stats. Every single word, even the smallest parts of speech, makes a big difference when defining data. Digital coupons being thought of as a convenient option does not mean all 76 percent of those shoppers would actually use digital coupons. The second statistic’s wording implies it would be the more reflective of the two of how important shoppers actually consider digital coupons to be.
Common Statistical Issues
If you are reading a statistical report and you see the words “significant”, be careful how you interpret it. Statistical significance and practical significance can be two very different concepts when it comes to your marketing department. A 1.6 percent difference may be significant through math-glasses, but it might not even be worth your consideration in practical terms.
Statistical reports may present you with correlation factors between each and every variable. But keep in mind that correlation does not always imply causation. There can be other variables in the mix linking the two, causing an indirect relationship, or the pattern can even be caused simply by coincidence.
You need to make sure any survey you conduct or use the results of employs a random sample — that is, a random representative group of what you could expect all of your prospects or targets to be like — or your results will reflect a sample bias. Not only does the group have to represent the population’s characteristic distributions, but it also has to be large enough of a sample size to get a reliable statistical reading. Generally, 25 data points is the accepted minimum for statistical reliability. But the more data points you have, the more representative your results will be of reality.
Then there is the response bias, which covers a wide array of cognitive or external pressures that may result in a person not responding to survey questions truthfully. Surveys are very easy to skew and their methodologies should be scrutinized carefully.
There are dozens of errors that can twist your results and decisions. The good news is that though there are so many, errors themselves are distributed normally, so they’re not wreaking complete havoc on our marketing campaigns. But some errors are outliers: Pay attention to the stats and methodologies you apply to your marketing strategy.