Digital Advertising and Firm Size

How to cite

Liaqat, Asad; Dubé, Jean-Pierre; Moshary, Sarah; Wernerfelt, Nils. 2022. "Digital Advertising and Firm Size." AEA RCT Registry. AEARCTR-0010260. https://doi.org/10.1257/rct.10260-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Asad Liaqat* , Jean-Pierre Dubé , Sarah Moshary , Nils Wernerfelt

* Primary investigator

Registry ID
AEARCTR-0010260
DOI
https://doi.org/10.1257/rct.10260-1.0
First registered
Last updated
Status
In development
Trial dates
2022-10-17 – 2023-06-30
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/10260

Abstract

We test whether the return to digital advertising differs by firm size using data from a large-scale randomized control trial on Facebook. We conduct thousands of experiments that randomly assign users to test and control conditions, where treatment users are eligible to see ads from the advertiser in question but control users are not. We then conduct a meta-analysis using the cross-section of advertising treatment effects to test for a systematic firm-size difference. Of interest is whether advertising effects are systematically larger for smaller firms. Our findings will shed light on whether digital advertising lowers the marketing barriers to entry that have facilitated market concentration in 20th century consumer goods markets.

Intervention

For each advertiser in our sample, we will randomly allocate a small fraction of their ad budget across users on the Facebook platform, including holding some users out from seeing the ads.

Experimental design

We will select a number of advertisers at random from Facebook who either answered our survey or are similar on observables, where we will stratify by firm size where applicable. For each advertiser in our sample, we will randomly allocate a small fraction of their ad budget across users on the Facebook platform, including holding some users out from seeing the ads. Using this sample of randomly-allocated ad spend, we will estimate the following equation for each advertiser:

Conversion_{ic} = a_c +b_c*AdExposure_{ic}+e_{ic}

Where i denotes the consumer and c denotes that company. The LHS variable will be measured both as a purchase indicator and using spending in USD where applicable.

We can then plot the estimated ad effects (b_c) against our measure of firm size (scatterplot). To test whether ad effects differ systematically by firm size, we then regress the estimated ad effect coefficients (b_c) on our measures of firm size:

\hat{b}_c = \alpha + \sum_{b=1}^B \beta_b * size_{bc} + \epsilon_{c}.

It will be important to correct for the fact that our dependent variable is a generated variable. We are interested in testing whether we can reject the null hypothesis that the coefficients on the size indicators are zero.

To allow a more flexible analysis and incorporate shrinkage, we may also explore using empirical Bayes methods (e.g., Efron (2016)) as other recent digital advertising studies have.

Design details

Randomization method
Randomization done using Facebook internal experimentation infrastructure.
Randomization unit
Advertiser
Clusters
N/A
Observations
8,000 advertisers
Arms
8,000 advertisers
Primary outcomes
Our primary outcomes of interest are cost per incremental customer and return on advertising spending.

Cited by

No citation data available for this registration yet.

Keywords