Will anybody buy? Logistic regression - Improving the User Experience through Practical Data Analytics - page 296

Next, move down the igure to the Classiication Table (middle section). It tells
you that this regression analysis predicts 87.7% of the case correctly (see oval in
Figure 11.16 ). Speciically, the vast majority of the people who answered “No” to
the question whose answer is our “Y” (Is there any kind of online news experience
that you would be willing to pay for?) are correctly predicted (90.6%), and the
vast majority of those who answered “Yes” were predicted correctly (83.7%—72
correctly predicted, 14 incorrectly predicted). As noted, overall, we have 87.7%
predicted correctly. This 87.7% fares very favorably with both the Cpro and
Cmax values. (See earlier sidebar on Cmax and Cpro.) There are 107 “0's” (No)
and 86 “1's” (Yes) as values of Y. The values are the row totals in Figure 11.16
(106 + 11 = 117 and 14 + 72 = 86). Thus, 117 is the frequency of the largest group:

Cmax=117/203=0.576=57.6%

andCpro= (117/203) 2 + (86/203) 2 =0.512=51.2%.

If a hypothesis test were performed to see if the 87.7% is statistically signiicantly
higher than the 57.6%, the result is signiicant with p -value < 0.001, leaving little
doubt that the equation classiies beyond any reasonable doubt better than random
chance, using even Cmax, the higher benchmark, as the baseline.
The third section of the output tells us an even richer story. By examining the
“Sig.” column, you can see that there are ive signiicant variables, i.e., ive of the
p -values are below 0.05. We've listed them with their respective p -value and coef-
icient sign in Table 11.4 .
OK, now you're ready to make some really powerful predictions based on empiri-
cal evidence. Since all the coeficients of these signiicant variables are positive, we
can conclude that, everything else being equal,
•

a person who is of the opinion that “unique local content not available on local
newspaper sites” is more impactful on his/her likelihood of paying for an online new
experience is indeed more likely to be willing to pay for an online news experience;

• a person who subscribes to a print newspaper is more likely to be willing to pay

for an online news experience;

• a person who is of the opinion that having Unique Local Content Not Available

on a Local Newspaper Site is more impactful in making a decision about his/her
likelihood of paying for an online news experience is more likely, indeed, to be
willing to pay for an online news experience;

Table 11.4 The Signiicant Variables in the Multiple Regression; CharlestonGlobe.com

• Education ( p -value = 0.002 + coeficient)

• p_newspaper ( p -value = 0.000 + coeficient)

• Local ( p -value = 0.011 + coeficient)

• Credibility ( p -value = 0.010 + coeficient)

• Business ( p -value = 0.000 + coeficient)