Does your analysis seem a bit lightweight at first glance? Do other UXers' charts and graphs kick sand in your deliverable's face?
Don't be afraid of numbers! You— yes, you!— can quantify anything!
You'll learn how to start getting real numbers out of your qualitative research, and get over the fear that you can't "do the science stuff". Plenty of rapid-fire real-world examples, including:
- Content audits!
- Web analytics!
- Bargain wines!
- Gendered social interaction in science fiction television shows!
...no, really!
29. Read “How to Think about Statistics” Write me with questions: wendy.stengel@gmail.com Twitter @wendywoowho Learn More
Notes de l'éditeur
Some of you are sitting here, shifting in your seats, filled with a thrill of mixed excitement and dread. The very word “statistics” can do that to a person – especially a person with, say, a B.A. instead of a B.S.If you want to make the mystical power of descriptive statistics work for you, instead of against you, you need to get past the dread.
Don’t fear the lingo. As UX pros, we’re no strangers to industry-specific jargon – we know our HTML from our CSS, we know our Dublin Core, SEO, SEM, CMS…the whole UX alphabet soup. There’s absolutely no reason why you can’t learn a few more wonky terms.
For example… Descriptive Statistics is a way to talk about data attributes. That’s not so scary – we talk about data all the time! In fact, when you’ve been talking about test results in your deliverables, you’ve already been using the most basic of descriptive statistics. The most basic is COUNTING. Just counting. How many bottles of wine are in the wine closet? 481. That’s a descriptive statistic!
When we start breaking it down by type of wine, we’re working with the next-most-basic type of statistic: the CROSS-TABULATION.Now, that may not seem very impressive to you – other than the fact that it’s a buttload of wine (or, rather, 4/5 buttloads) – because it’s “just a table.” You can make it in Word, you can make it in Excel…it is EASY.And, yes, it’s STATISTICS. A STATISTICIAN would call it a CONTINGENCY TABLE. It is a MATRIX DISPLAY of the categories of two NOMINALLY SCALED VARIABLES.I slipped a few more terms in there, didn’t I? A VARIABLE is a VALUE. It can be a QUANTITY or a QUALITY.With our wine example, “red” is a variable – it’s a quality of the wine.
NOMINAL SCALE variables are CATEGORICAL variables. I’m an IA – we’re all about categories.
So, for our wine closet, we can categorize wine by type, and by style. The “stuff” we put in those categories isn’t numbers – we’re putting in adjectives here. There’s no inherent order to them – red doesn’t have to come before white, sparkling doesn’t have to come after standard. They are just names, hence “nominal.”But some types of variables DO have inherent order to them. There’s a couple different types.
ORDINAL SCALE VARIABLES are variables that have a RANK attached to them.We’re ranking experts!
In a usability test, we rank task completion: Succeeded, Succeeded with Effort, and Did Not Succeed.Effortless success is better than sweaty, painstaking success, and sweaty, painstaking success is better than failure. We can talk about that with ORDINAL VARIABLES, but we can’t talk about how MUCH better one is than the other. There’s no meaningful distance between the ranks.
To talk about meaningful distances, we need INTERVAL VARIABLES.INTERVAL VARIABLES are probably the sorts of things you think about when you think about STATISTICS. They are NUMBERS, folks.
Let’s play a little game I like to call Spot The Variable:* What kind of variable is shown by the little wine glasses on the left? NOMINAL* What kind of variable is shown by the scores on the right?* What kind of variable is shown by the value/price?Great! Now we’re all limbered up, we don’t fear the lingo.It’s time for NUMBERS.
Don’t fear the numbers.Numbers are everywhere. You deal with numbers every day.
Where ever you see a pattern, you’re seeing numbers. The grid you use to lay out a pageThe lacy bits on a camisoleIf you look for the numbers, you’ll find them. Because, here’s one of the big mysteries of statistics…
YOU can quantify ANYTHING.Let’s say that again: YOU – yes, YOU– can quantify ANYTHING.
Take, for example, gendered social interaction in science fiction and fantasy television.Now, I’ve got a BA in literature, and concentrated on popular culture. I spent a lot of time analyzing Star Trek, and argued that, Friendship is one of the most highly prized things in the Trek universe…and that it was largely reserved for men. In the original series, male – male friendships are everywhere. In Next Generation, men and women are friends with each other, which is a great addition, but the friendships shown between women are shallow at best. I argued this with all the tools in my literary gender theory toolkit.It was pretty darned sweet. But…how could I prove to my science-minded friends that what I said was …TRUE. Valid. Representing a complete set of available data.Well, the only way to do it was to get objective data. I started looking at what specific, measurable things happened in gendered social interactions.Calling people by their name. Calling people by a nick name.Touching people.Going to bed with people.Eating, drinking, enjoying hobbies, playing poker. Everything became a variable. Does X happen in this scene? Who wrote this episode? Who directed it? I ended up with many, many, many data sheets, and a heck of a data base.Everything was a number. Every. Little. Thing. It was a lot of hard work.
In your practice, you are ALREADY DOING THE HARD WORK.Say you’re doing a content audit. You’ve got a SPREADSHEET OF DOOM.
Is it a Spreadsheet of Doom? Well, yes, but it’s also a STATISTICAL BEACHHEAD.It just needs a little nudge.
First, you’ve got to assign numbers to your data.Here, I’ve put in a 5 point numerical scale to match the 5 point content score. I’ve also numbered the audiences, and each audiences goals.
THIS IS COMPLETELY LEGITIMATE.Once you have numbers assigned to your variables, it’s time to do the science stuff.
DO NOT FREAK OUT!You CAN do the science stuff. When you do user research, you’re doing social science, and social scientists have every bit as much need of statistics as the “hard scientists.”And what they do, we do – we can…
REJECT THE NULL HYPOTHESIS.That is, to get sound information about relationships between variables, we have to assume that they are NOT related, and then test to see if that is true.There are three different tests we can use to figure that out, and if you choose the right test, you’re in like Flynn.
Remember those bits of jargon I said you didn’t need yet? Now you need them.You don’t have to do any long hand. You don’t have to punch cards and feed ‘em in a computer like my mom did when she took statistics. You don’t even need a powerhouse statistical program for most of the tests you’re likely to run. These three tests? ARE BUILT IN TO EXCEL. You’re just a click away from correlational bliss.But you need to choose the right test. Take the two variables you want to prove aren’t related, and identify their scales. If you’re nominal/nominal, use a Chi-Square. If they’re nominal/interval, ANOVA. If they’re interval/interval, you can use a much more powerful test…Pearson’s.You’ll notice that “ordinal” doesn’t show up on that list. A hard scientist would likely treat ordinal scale data as nominal.Much of what social scientist deal with, though, is ordinal scale data, like the content score in our little content audit. How can we get the most powerful analysis out of ordinal scale data?I’m going to give you the key to happy social statistics:
Treat your Ordinal Scale Data Think of any opinion poll results you’ve heard – they’re using numbers, and those numbers came from 100% ordinal scale data – how likely are you to… how would you rate…as Interval Scale Data.
THIS IS PERFECTLY VALID. Ordinal Scale Data can be treated as Interval Scale Data.So, you’ve run your appropriate test, and you’ve been able to reject the null hypothesis. You have a correlation! You’re ready to go tell your client that the presence of a link to your privacy policies in your footer correlates to purchases of widgets!….hold your horses.
CORRELATION DOES NOT EQUAL CAUSATION.Taking Algebra 2 in high school correlates strongly with success in your post-college work. The correlation is there.But taking Algebra 2 doesn’t CAUSE it. Perhaps those who take Algebra 2 are more likely to be interested in challenging themselves. So, you need to be …sensible. When you have a result, be happy, but then THINK about it.
Be thoughtful.Be sensible.Seek truth.You are all very smart, very curious people, who give a damn about their work. You wouldn’t be here otherwise. So trust in the fact that your brain, your common sense, and your commitment to truthfully representing things will help you when you’re starting out with statistics.
This WILL help your practice.
If you test your assumptions, your recommendations will have firmer footing. You’ve seen it with web analytics…slicing and dicing the numbers gives us more gravitas.This will reassure your clients.This will boost your confidence.