Subscribe to DSC Newsletter

Robust Binomial Proportion Interval Population Estimates

Hi,

I have a situation where I have 100 reports available on Direct Marketing campaign.

These above all are intended to measure success of the treatment over control.

However the control & treatment group N's are too small for the marginal effect sizes of DM campaigns to generate a reasonable width interval for evaluation with a standard normal approximiation for binomial proportions.

 

What i'd like to know...what is the industry standard for using alternatives like bootstrapping, wald intervals to be able to create a meaningful interval for testing the above scenario.

As an example:

Treatment Nt = 1000

Treatment pt = 0.5%

Control Nc = 100

Control pc = 0.7%

 

 Regards,

Clancy.

Views: 622

Reply to This

Replies to This Discussion

Hi Clancy,

Not sure if this applies to your setting, but I designed a non parametric methodology for confidence intervals. See my article at http://www.analyticbridge.com/profiles/blogs/how-to-build-simple-ac....

Best,

Vincent

Hi Clancy,

Did the post below provide some help to answer you?

Cheers,

Tim 

Thanks Vincent. Interesting approach. Seems similar to bootstrapping.


Vincent Granville said:

Hi Clancy,

Not sure if this applies to your setting, but I designed a non parametric methodology for confidence intervals. See my article at http://www.analyticbridge.com/profiles/blogs/how-to-build-simple-ac....

Best,

Vincent

Hi Tim, 

I read the article posted by Vincent and got something out of the approach.

However I ended up using bootstrapping to approximate the sampling distribution of the mean for my response rates.

Regards,

Clancy.


Tim Madison said:

Hi Clancy,

Did the post below provide some help to answer you?

Cheers,

Tim 

I'm happy to hear that you found a solution.  Thanks for using DSC and participating in the community

Hi Tim, 

I definitely get value from being a member of this other data science related groups via linked in etc...

I think there is so much to gain for many in the industry by talking and communicating. Especially at this time when there is a shortage of skills in this space. Especially felt within my community in Brisbane Australia.

The hardest battle I will come up against time and again however will not by technical but rather convincing and showing the business' I consult for a) there is massive value in data science for their organisation and b) what the outputs of the activity are that they can get the most use from.

Also the strength of community is important in a growing industry. 

Regards,

Clancy.

Thanks Clancy - keep up the good fight :-)

CIs for proportions, and differences between proportions, is one of the most studied problems in all of statistics.  An entire literature examines the problem under conditions of small sample sizes.  Cites to several open access, authoritative (arguably seminal) survey papers are below.  Great papers I’ve been able to use under a wide range of data conditions.  Best, J.D.

http://projecteuclid.org/DPubS/Repository/1.0/Disseminate?view=body...

http://projecteuclid.org/DPubS/Repository/1.0/Disseminate?view=body...

http://www.jstor.org/pss/2685469

 

I think it is very interesting that a group formed around 'big data' discusses small sample issues :-)

I think it says something about datascience in general.

Not sure what it is though ;-)

Thanks. Keep going

RSS

Videos

  • Add Videos
  • View All

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service