Estimating The Causal Effect Of Online Display Advertising

  • August 11, 2011 · 7:00 PM

We are excited to have Ori Stitelman of Media6Degrees join us in August to discuss his important work on estimating causal effects of display advertising.  Ori's abstract & bio follows...

ABSTRACT- This talk will examine ways to estimate the causal effect of display advertising on browser post-view conversion (i.e. visiting the site after viewing the ad rather than clicking on the ad to get to the site). The effectiveness of online display ads beyond simple click-through evaluation is not well established in the literature. Are the high conversion rates seen for subsets of browsers the result of choosing to display ads to a group that has a naturally higher tendency to convert, or does the advertisement itself cause an additional lift? How does showing an ad to different segments of the population affect their tendencies to take a specific action, or convert? We present an approach for assessing the effect of display advertising on customer conversion that does not require the cumbersome and expensive setup of a controlled experiment, but rather uses the observed events in a regular campaign setting. Our general approach can be applied to many additional types of causal questions in display advertising.

BIO-Prior to joining Media6Degrees, Ori received a Ph.D. in Biostatistics from the University of California, Berkeley. His research concentrated on developing methods for estimating causal effects. In particular, he focused on estimating causal effects in observational, censored, and longitudinal data. At UC Berkeley, Ori and his advisor, Dr. Mark van der Laan, focused on Targeted Maximum Likelihood Estimation (TMLE), a class of doubly robust semi-parametric estimators that focus on estimating a causal parameter that directly answers a business or scientific question of interest. In addition to publishing several papers on the topic, Ori recently contributed two chapters to the first book published on TMLE, "Targeted Learning: Causal Inference for Observational and Experimental Data" by Mark van der Laan and Sherri Rose. To date, TMLE has been primarily applied to health outcomes data. However, the causal methods he and his group developed while at UC Berkeley are widely applicable across many different subject areas. Currently, Ori is working with his colleagues at Media6Degrees to implement those same methods to estimate the causal effect of display advertising on browser conversion.

Join or login to comment.

  • Omer C.

    Topic was quite good, but I think the speaker could have done more to illustrate his work in a business setting - I felt he was more focused on theory vs application.

    August 20, 2011

  • A former member
    A former member

    An excellent lecture in a very nice environment.
    I am looking forward to the slides that Ori promised to post on meetup and on his blog (on the 5 degrees of connection website).

    August 12, 2011

  • Sharon C.

    Good/interesting presentation on an alternative to A/B testing. However, I'd like to better understand how this method manages to control for "all confounders" - do the possible confounders not need to be defined beforehand (in which case it's easy to miss out on something)? In addition, advertisers looking to do say, a view through test, on their own would still need to run PSAs just to collect data on non-exposure, so the point about not needing to spend $ running PSAs wouldn't necessarily apply to everyone. Looking forward to the deck/paper.

    August 12, 2011

  • A former member
    A former member

    Lecture was interesting, but the foundations of the study - what exactly was checked, what customers were targeted, and mainly - the companies and the specific websites that the companies' ads were placed on - were not explained well at the beginning and caused a lot of confusion. It could have been also useful to show a similar study but for companies that are not well known. I assume that the three telecom companies that were tested are companies consumers are well aware of. Picking up less known companies might have shown different results, also between the various models that were examined in the study.

    August 12, 2011

  • Gustavo C.

    Really in the initial stages; wasn't clear what changed in the selected group to bump up the results or compare this to the "naive" approach of simply selecting the "hot" viewers. What is the advantage of using this blended / filtered / averaged model ?
    Also assumes you can't hand-tweek the existing ad hoc models for better results

    August 12, 2011

  • A former member
    A former member

    A very enjoyable and comprehensive presentation! Looking forward to reading the paper.

    August 12, 2011

People in this
Meetup are also in:

Create your own Meetup Group

Get started Learn more
Allison

Meetup has allowed me to meet people I wouldn't have met naturally - they're totally different than me.

Allison, started Women's Adventure Travel

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy