Can you relate in multiple ways? Multiple linear regression and stepwise regression - Improving the User Experience through Practical Data Analytics - page 267

recruiters used the search for a while, we asked each participant to rank their percep-
tion of usefulness of the different search ields, along with their likelihood to adopt
the search engine.”
“Okay,” he says slowly, riling through the deck. “Keep going, I'm listening.”
“Then we analyzed the data, and determined which combination of search
ields positively affected the likelihood of adoption of the search engine they had
just used.”
“Really?” he says, looking over his reading glasses. “Sounds like magic.”
“No, just stepwise multiple regression.”
Velluci look up at Hans. “Do you follow any of this?”
“Hmm…not exactly. My college stats class was a long time ago.”
“Yeah, me too,” he agrees. “But I have some questions.”
“Sure,” you reply with a grin.
“What's your sample size?”
“180.”
“All bonaide professional recruiters who had never used our current search?”
“Yes.”
“And they rated Boolean as the most useful ield?”
“Yes, it had the highest positive correlation with likelihood of adoption.”
“OK…,” he says. “And where is this so-called ideal search engine you've been
able to identify?”
“It's on page 11. Basically, we have the most important ields on the Basic Search,
with the less important ields on the Advanced Search.”
“How can 'advanced' be less important?” he says, with more than a hint of sar-
casm, as he studies the design.
“Good point,” you reply calmly. “It's been standard practice on the Web for a
while. You call esoteric ields, “Advanced,” and hide them in a pop-up. It sounds
better than, “Esoteric.”
“You got that right,” he chuckles. But he turns more serious. “How conident are
you of these indings?”
“Well, you always need a sample size of at least 10 per independent variable to
have conidence in the stepwise regression results. Since there were 15 ields, we
were more than covered with 180. And all the p -values of the ields we've included
in the Basic Search are way below 0.05.”
Joey uses the under-the-throat “cut” sign. “Answer the question.”
“Yes,” you state.
“Good”, he states. “It's refreshing when I get a straight answer around this place
once in a while.” He pauses, then holds up your report in one hand while pointing to
it with his other hand. “But let me ask you something.”
“What's that?” Hans asks, trying to stay relevant in a conversation that is above
his head. Velluci ignores him and stares at you.
“Why didn't one of my clown VPs ask UX to do this kind of great research before
I spent a boatload?”