Observational Learning: Evidence from a Randomized Natural Field Experiment

ISPS Data Archive: Terms of Use

By using, contributing, and/or downloading files associated with scholarly studies available on the ISPS Data Archive, you agree to these terms and conditions. Please read the ISPS Data Archive Terms of Use.

Suggested citation: 

Cai, Hongbin, Yuyu Chen, Hanming Fang (2009). Replication Materials for: ‘Observational Learning: Evidence from a Randomized Natural Field Experiment,’ http://hdl.handle.net/10079/tx95xkj. ISPS Data Archive.


Hongbin Cai, Hanming Fang, Yuyu Chen

Research design: 
Data type: 
Data source(s): 


Data source information: 


Field date: 
October 16, 2006
Field Date: 
2006-10-16 - 2006-10-30
Location details: 
Asia - China
Unit of observation: 
Dining party (bill)
Sample size: 
We used 800 CNY (Chinese Yuan) as the cutoff, above which the bill was considered large. From a total of 13,302 bills in our dataset (including both the pre-experiment and experiment periods), a total of 407 bills were deleted due to these considerations. The deletion of these large bills affects only the calculations of the means for dishes ordered and bill amount, but does not at all affect subsequent analysis of the effect of observational learning on customers’ choices. Including these large bills would lead to significantly larger means for both dishes ordered and the bill amount, inconsistent with what the restaurant managers would consider reasonable.
Randomization procedure: 
Diners are randomly assigned to tables with three different information conditions. We implemented a two-stage randomization strategy where the first-stage randomization was at the level of restaurant sites, and the second stage was at the level of tables within a site. Specifically, in the first stage, we randomly selected five locations where tables in each location were subsequently randomized into control tables and ranking treatment tables; and we randomly selected four other locations where tables in each location were subsequently randomized into control tables and saliency treatment tables. After randomly selecting the locations for the ranking and saliency treatments (the first-stage randomization), we randomly assigned tables in each of the selected locations into control and treatment tables (the second-stage randomization).
The first group of tables are the “control” tables where no additional information about the dishes other than the regular menu is displayed on the tables. The second group are the “ranking treatment” tables where we place a 19 by 12 centimeter plastic plaque on the table displaying the names of the five most popular dishes sorted by the actual number of plates sold in the previous week in that location, with the number-one dish listed on top. The actual numbers of plates sold are not displayed. Note that the top 5 rankings may differ by locations. The third group of tables are called the “saliency treatment” tables where we place a same-size plastic plaque on the table displaying the names of five sample dishes from the menu, sequenced in a random order.a
Treatment administration: 
19 by 12 centimeter plastic plaque.
Outcome measures: 
Demand for top 5 dishes
Archive date: 
Owner contact: 


Terms of use: 

ISPS Data Archive: Terms of Use

Data file number Description File formatsort ascending Size File url
D050F02 Dataset - main Stata (10.0) .dta 113246208 Download file
D050F07 Program file Stata (10.0) .do 4300 Download file
D050F06 Program file Stata (10.0) .do 6963 Download file
D050F05 Dataset - summary stats Excel .csv 355328 Download file
D050F04 Dataset - main Excel .csv 100663296 Download file
D050F01 ReadMe file Adobe acrobat (8.0) .pdf 33792 Download file
D050F15 Metadata (DDI 3.2) .xml 235730 Download file
D155F06 Program file - analysis script .r 13810 Download file
D050F08 Program file .R 21431 Download file
D050F09 Output .log 285696 Download file
D050F10 Output .log 37888 Download file
D050F03 Dataset - summary stats .dta 414598 Download file