Cox’s Proportional Hazard Modelling

[Question]

Hi All,

The objective is to build a survival model..Predicting when customers are more likely to return and purchase...For example, we are expecting the customer 101 to return on w9 etc.The data below represents the weekly customer transactions they have made in the last 13 weeks...(the value in the data is the amount spent in that week).

Hope It makes sense, if not please do not hesitate to give me a shout.

Thank You

customer_id	W1	W2	W3	W4	W5	W6	W7	W8	W9	W10	W11	W12	W13
100	.	.	.	.	.	.	2.5	.	.	2.5	2.5	.	.
101	4.5	2	3.5	2.5	2.5	2.5	.	.	4.5	5	.	.	.
102	.	6.5	.	.	.	.	.	.	6.5	.	.	9	.
103	.	5	10	.	.	.	9	8.5	13	.	7	18	7
104	2	2	.	.	.	.	.	.	2	.	4.5	2	.

[Question]

Hope It makes sense, if not please do not hesitate to give me a shout.

Thank You

customer_id	W1	W2	W3	W4	W5	W6	W7	W8	W9	W10	W11	W12	W13
100	.	.	.	.	.	.	2.5	.	.	2.5	2.5	.	.
101	4.5	2	3.5	2.5	2.5	2.5	.	.	4.5	5	.	.	.
102	.	6.5	.	.	.	.	.	.	6.5	.	.	9	.
103	.	5	10	.	.	.	9	8.5	13	.	7	18	7
104	2	2	.	.	.	.	.	.	2	.	4.5	2	.

[Question]

Hello All,

I would appreciate if you could answer my message below..

Thank You

[JacobSimonsen]

I think the Cox model is not the right model to use when the goal is to predict the event time. This is because of the non-parametric baseline hazard rate, that make it difficult to translate the estimated hazardratios to predicted eventtimes. This can be seen by the fact that the Cox model doesnt care about when things happens, but only in what order things happpens.

Then better use the full parametric accelrated failure time model (for example the weibull model). For this you can use PROC LIFEREG.

[Question]

Thank You Jacob for your response...Are there any white papers. documents
and example on weibull model you could recommend? I have never built one so
it would be good to get some documents explaining how it works...

Cheers

##- Please type your reply above this line. Simple formatting, no
attachments. -##

[JacobSimonsen]

There is a very good documentation in the SAS help. And also some good examples. As far as I remember, you can use the output statement, and there get predicted values.