Logistic回归用于R中的广告分析

Question

I have a data set where I am trying to test if having a 7 day advertising period is better than having a 5 day advertising period. I feel that logistic regression would be the best way to test this out. I ran the tests for 2 weeks each. I have data like traffic, signups, attrition.

Here is what the data look like:

              5d         7d  greater (does the 7d have atleast 5% more than 5d)
Traffic     179650  196395   1
subscribers 437899  442068   0
attrition   2304    2376     0
signups     5039    6246     1

1 means yes and 0 means no.

I ran this code in R:

fit2<-glm(greater~X5d + X7d, data=logr2, family = "binomial")

then

predict(fit2, data=logr2, type = "response")

My output is :

 1            2            3            4 
1.000000e+00 6.753019e-13 1.386707e-10 1.000000e+00 

or

> round(predict(fit2, data=logr2, type = "response"))
1 2 3 4 
1 0 0 1 

How can I run it such that I get only 1 output to tell me 1 or 0 (IE does the 7 day have a greater than 5 % overall increase?)

Thanks

Answer 1

I think you've confused the argument names of the predict function (see the documentation ), try this:

predict(fit2, newdata=logr2, type = "response")

The strange output comes from the fact that you're giving the training data as the input to predict, which doesn't really make sense. Try some fresh data point, like this:

input = data.frame(X5d = 123, X7d = 22)
predict(fit2, newdata=logr2, type = "response")

Result:

1

2.775557e-13

which means that it's 1 with probability almost 0 .

If you give an exact point from your data set:

input = data.frame(X5d = 179650, X7d = 196395)
predict(fit2, newdata=input, type = "response")

Result:

1

1

So it's 1 with probability 1 .

You can check other data points from your training set - the results are perfect, as for such a few training data samples your fit is ideal.

You can find a simple, similar example here .