Reducing fraud through pre- or post- control

Last registered on March 22, 2021
View Trial History

Pre-Trial

Trial Information

General Information

Title
Reducing fraud through pre- or post- control
RCT ID
AEARCTR-0007015
Initial registration date
January 13, 2021

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
January 14, 2021, 11:57 AM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated
March 22, 2021, 10:13 AM EDT

Last updated is the most recent time when changes to the trial's registration were published.

Locations

Country
France
Region
Europe

Primary Investigator

Name
Francisco de Asis Martinez Jerez
Affiliation
The Hotel School, Cornell SC Johnson College of Business, Cornell University
Contact Primary Investigator

Other Primary Investigator(s)

PI Name
Pablo Casas-Arce
PI Affiliation
Arizona State University
Contact Investigator

Additional Trial Information

Status
In development
Start date
2021-03-22
End date
2022-09-22
Keywords
Behavior, Firms & Productivity, Labor
Additional Keywords
control; fraud
JEL code(s)
D82; J44; L14; M55
Secondary IDs
Abstract
We are testing in the field the relative efficacy of controlling the costs of a job with an ex-ante third party assessment of the cost of the job versus an ex-post revision of estimates that seem inflated.
External Link(s)

Registration Citation

Citation
Casas-Arce, Pablo and Francisco de Asis Martinez Jerez. 2021. "Reducing fraud through pre- or post- control." AEA RCT Registry. March 22. https://doi.org/10.1257/rct.7015-1.1
Sponsors & Partners
Experimental Details

Interventions

Intervention(s)
Intervention Start Date
2021-03-22
Intervention End Date
2021-06-22

Primary Outcomes

Primary Outcomes (end points)
Repair estimate by the professional
Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)
Time to perform the repair
Secondary Outcomes (explanation)
Time elapsed from the time the customer calls to notify the damage until the repair is finalized and accepted by the customer.

Experimental Design

Experimental Design
In collaboration with a leading French business process outsourcer of repairs for property and casualty insurance we engineered a field
experiment by randomizing whether the professional executing the repair received an ex-ante estimate of the cost of the repair or not.
Experimental Design Details
The company is a business process outsourcer of repairs for property and casualty insurance. In the current process, an independent assessor first estimates the cost of the repair and then the repair is offered to a professional (painter, mason, etc.) who makes her own estimate of the cost of the repair that is approved or not by the insurance company.
The company is testing the substitution of the first assessor (FIA, first intervening agent) by an automated algorithm fed with the information provided by the insured person.
We are testing the efficacy of the automated algorithm in curbing fraud (inflated estimates) by the professional. The estimate by the automated algorithm will be used in two forms: (1) without communicating to the professional that there is an automated estimate and checking ex-post the difference between the estimate of the algorithm and that of the professional to identify potential fraud; and (2) communicating to the professional the estimate of the automated algorithm in the same form that today the professional receives the estimate of the FIA.
Therefore there are three interventions:
1.- Control group: an independent assessor first estimates the cost of the repair and then the repair is offered to a professional (painter, mason, etc.) who makes her own estimate of the cost of the repair. In this case the professional knows the estimate of the independent assessor.
2.- No ex-ante estimate group: the repair is offered to a professional (painter, mason, etc.) who makes her own estimate of the cost of the repair. In this case the professional does not receive any ex-ante estimate. Unknown to the professional, during the trial, an independent assessor will estimate the cost of the repair after the professional has issued her estimate. An ex-ante algorithm-based estimate of the repair will be compared to the professional estimate to test the ability of the algorithm to identify estimate inflation.
3.- Algorithmic ex-ante estimate group: An algorithmic program produces an ex-ante estimate of the repair based on customer provided information. The repair is offered to a professional (painter, mason, etc.) who makes her own estimate of the cost of the repair. In this case the professional knows the estimate of the algorithm. Unknown to the professional, during the trial, an independent assessor will estimate the cost of the repair after the professional has issued her estimate.
We randomly assign repairs to each of these treatments.
Randomization Method
Claims arrive randomly to the phone bank of the company.
We assigned them to treatments in two stages: first, by day of arrival and, second, by repair ID.
(1) Because of limitations in the operating system of the company we had to assign all the repairs of one day either to the control group or to the treatment groups. We flipped a coin to decide on the assignment for the first day. Then we switch daily between treatment and control so there was an even distribution among both groups in terms of day of the week, beginning and end of the month, and seasonality.
(2) All repairs were assigned an ID number. The days of treatment repairs were assigned to treatment a or b as a function of whether the repair ID number was even or odd.
Randomization Unit
The randomization unit was the repair
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
Approximately 1,800 repairs
Sample size: planned number of observations
Approximately 1,800 repairs
Sample size (or number of clusters) by treatment arms
Approximately 900 control, 450 treatment a, and 450 treatment b
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
10% differences in estimates. Disguised numbers $400 average repair, $240 standard deviation, $40 minimum detectable effect at 5%
Supporting Documents and Materials
IRB

Institutional Review Boards (IRBs)

IRB Name
IRB Approval Date
IRB Approval Number
Analysis Plan

Post-Trial

Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Intervention

Is the intervention completed?
No
Data Collection Complete
Data Publication

Data Publication

Is public data available?
No

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials