Steering in Online Markets:

The Role of Platform Incentives and Credibility

http://john-joseph-horton.com/papers/badging.pdf

​

Moshe Barach, John Horton (@johnjhorton), Joe Golden

http://bit.ly/2M0i64r

Platform markets have preferences over matches that are formed

Platform markets have preferences over the matches that are formed on the platform

• Direct revenue might differ for different sellers
• Platform "take" differs
• Customer service costs
• Bad experiences spill-over on other sellers
• Repeat business (Jaffe et al. "Airbnb paper"; Nosko & Tadelis "eBay paper")

​

Platform markets tools to steer buyers and/or make transactions more likely

Platform explicit (algorithmic)

recommendations

Expose or highlight product "facts" / reviews

Auction off visibility (hope for pos. selection)

Offer incentives

(for example, money-backed guarantees)

Money-backed guarantees

• Pro:
• Reduces the financial risk to buyers; platform better able to absorb risk
• Cheaper for the platform to guarantee the "best" sellers, which in turn might make these implicit recommendations more credible
• Con:
• Might encourage moral hazard in selection
• Could be expensive

This paper

• Report the results of two steering experiments in an online labor market
• Main result: Money-backed guarantees strongly affect which buyer is selected, but don't increase matches overall & seem to work just as well as (non-financial) recommendations.

Who gets "guaranteed"?

Suppose a would-be employer (buyer) receives applications from some number of workers (sellers):

Score = f(WageBid, Stars, Hours-Worked)

Score

0.95

0.72

0.45

0.42

0.23

0.10

Score

0.95

0.72

0.45

0.42

0.23

0.10

Platform picks some cutoff score, with all sellers above that being "preferred"

Score

0.95

0.72

0.45

0.42

0.23

0.10

"Molly" and "Ada" would be "preferred" because they have scores above 0.5

First experiment

Employer posts job opening

Opening gets applicants; Platform scores

Ⓡ

Score > 0.5

"Guaranteed"

Status quo

Guaranteed

Status quo (not marked)

Sample size

Terms of the guarantee

Guaranteed sellers are really, really positively selected

> 50% more hours-worked

> 100% more past jobs

~ 50% higher wage bids

"Same" guarantee-eligible

sellers across two groups?

No difference in Guarantee-eligible sellers by buyer treatment status

Did offering the guarantee increase the number of buyers willing to hire?

No discernable increase in matches formed

Or in revenue conditional upon match being formed

No evidence of better matches

Uptake of the guarantee

Perhaps the guarantee was not salient and had no effect on buyers?

Even in the control, positive relationship between hiring & score

Above the cutoff, guaranteed sellers more likely to be hired

Some evidence for guarantee crowd-out

At the application level, we can interact I(score > 0.5) with applied-to opening treatment indicator

We can include a seller-specific FE in application level analysis

Being above threshold effect doubles when applying to an MBG opening

Some evidence of crowd-out---being below threshold and applying to an MBG job "hurts"

An ex ante concern was that buyers with guarantees would be less price sensitive

Higher wage bids; lower hire probability

No evidence treated employers less price sensitive

Could the same effects on selection have been achieved without the guarantee?

Employer posts job opening

Opening gets applicants; Platform scores

Ⓡ

Score > 0.5

"Guaranteed"

Score > 0.5

"Recommended"

Sellers who would have been guaranteed are "recommended"

No difference in probability contract formed

No apparent difference in selection

Comparing the two experiments side by side

Concluding thoughts

• Platform recommendations alone were powerful
• Buyers trust the platform
• Financial risks of hiring do not seem to be important
• Labor market implications (explains lack of bonds?)
• Ideas for future work
• Seller-provided guarantees