A bunch of young women are sitting around a table full of martinis talking about what it's like to be women in a male-dominated company. Theories for survival are being tossed about as casually as long locks of hair over bare shoulders. Then someone blurts out "It's gotta be more fun if you're blonde!" The small group falls silent as they wait for the only blonde at the table to address this assertion on behalf of all blondes, everywhere. She throws back the remainder of her apple martini, signals the waiter for another, and coolly declares "It doesn't matter. We can't ever really know for sure".
What is she talking about and what on earth does this have to do with marketing? In one word: causation. Marketers everywhere are tasked with finding out about causation. 'Is this advertising causing people to buy more of our lemonade?' 'Is our new viral marketing effort causing our brand awareness to increase?' 'Would having blonde hair cause me to have more fun…?'
It's virtually impossible to answer questions about causality with absolute certainty. But we can use techniques grounded in science to help us gather strong evidence of causation. Let's talk about designing an experiment to test whether or not blondes do, indeed, have more fun...
What Causes What?
The purpose of any scientifically designed experiment is to test whether the independent variable affects or causes the dependent variable. Now, which is which? In our experiment, the independent variable is the "blondeness," while the dependent variable would be the amount of fun being had. So the question we're trying to answer is: does the independent variable, being blonde, cause or affect the dependent variable, having more fun?
A non-marketer might try this experiment as follows. Take two single women, one blonde, one brunette, and add up the number of 'fun points' each attains during a one-week period. Let's say fun points are earned by attending social events such as parties, evenings out at bars and restaurants, concerts, charity fundraisers, and the like. At the end of a week, tally up the points and if the blonde gets more, we can conclude that blondes really do have more fun!
What's wrong with that? Lots. For starters, we have two different women with two different personalities and penchants for 'fun-seeking.'
Also, one woman could live in a notoriously 'fun' area such as Manhattan or Chicago and the other could live on the outskirts of a small town where opportunities to earn 'fun points' as they're defined here are slim and none.
What About Alternative Explanations?
But even if we hold the locations constant by choosing two women from San Diego, we've still got problems. One woman could hold down two jobs because she's saving for a house while the other is a notorious bar fly. Clearly there are a multitude of other variables - in addition to hair color -- that could also cause differences in how much fun they had.
Did the final tallies have anything to do with being blonde? Maybe, but we'll never know, because the design of our experiment did not allow us to rule out all of the other confounding variables. Our sample women were too different on too many fronts to be able to single out hair color as the cause of racking up 'fun points.'
Does this mean that we can never find out if blondes have more fun? Not necessarily. The lesson here is that when testing for causation, you must eliminate the differences between the samples being compared except for the one or two things you want to manipulate.
Doing it Right
Let's walk through a different, more marketing-related example of how to do it right. Let's say you're trying to justify spending extra money to upgrade from a black and white self-mailer coupon to a color coupon. So you want to test whether using color affects the number of coupons that are noticed and redeemed (or, said another way, 'whether the use of color causes coupon redemption to increase'). Your goal in designing this experiment is to come as close as possible to eliminating differences between samples, so that the only change that could account for increases (or decreases) in coupon redemption would be the color of your mailer.
How would you do that? Take your sample from a rather homogeneous population. Research shows that people in the same zip code tend to have lots in common in terms of basic demographics like age, income, ethnicity, and lifestyle. Choose a zip code and then take two random, equal-sized samples of coupon recipients. Perhaps you could narrow it down further to a particular neighborhood in that zip code, and deliver the black and white coupon to every other house, and the color coupon to those in between. Then count up how many of each enrolls after a certain period of time. If the averages of the two groups are significantly different, with the color coupons coming out ahead, you have some solid support for your theory that using color helps increase redemption rate.
Is it perfect? No. You can't ever eliminate 100% of the variability from human to human, but here's what's good about this design:
- You've selected a group of people, rather than just one person in each group, which will help to equalize differences among groups.
- You've selected homogeneous populations in terms of income, proximity to store location, lifestyles, etc, so purchases and coupon redemption should be pretty much the same in both groups (thus ruling out alternative explanations for the differences in redemption rates).
- You've selected people in the same area, so they will experience many of the same outside factors that may affect coupon redemption (weather, a parade or festival in town, etc.)
The main idea here is that we've done all we can to minimize other independent variables (weather, location, income, etc.) that differ among the two samples of coupon redeemers and that may affect the final tallies. The best remaining independent variable that differs between the two samples and that can go a long way to explaining differences in coupon redemption are the color of the coupon mailer.
So, Do Blonds Have More Fun?
So let's go back to our blonde question. How can we find out if blondes have more fun? Perhaps the best way to control for all other factors would be to use a one-woman sample. Put her on a nearly identical schedule for two consecutive weeks - one week as a blonde and one week as a brunette - and tally up her fun points at the end of each week and compare.
This time you've eliminated all personality, time-constraint, and location differences because it's the same woman. But now your biggest problem is finding that magical woman who looks as good as a blonde as she does as a brunette, or vice versa. If you think this is a simple task, I ask you this: have you seen Winona Ryder with blonde hair?
