Rivals in the Dark

Phillips et al. (2014): Rivals in the Dark: How competitition influences search and decisions under uncertainty

This package details the publication Rivals in the Dark: How competition influences search and decisions under uncertainty published by Phillips et al. (2014). All raw data and (most) analyses are contained in this package.

Installing the phillips2014rivals package

The phillips2014rivals package can be installed in R with the following commands:

# Install the phillips2014rivals package from GitHub

install.packages("https://github.com/ndphillips/ndphillips.github.io/blob/master/misc/phillips2014rivals_0.1.0.tar.gz?raw=true", 
                 repos = NULL, 
                 type = "source")

Abstract

In choices between uncertain options, information search can increase the chances of distinguishing good from bad options. However, many choices are made in the presence of other choosers who may seize the better option while one is still engaged in search. How long do (and should) people search before choosing between uncertain options in the presence of such competition? To address this question, we introduce a new experimental paradigm called the competitive sampling game. We use both simulation and empirical data to compare search and choice between competitive and solitary environments. Simulation results show that minimal search is adaptive when one expects competitors to choose quickly or is uncertain about how long competitors will search. Descriptively, we observe that competition drastically reduces information search prior to choice.

Vignettes (Guides)

The following vignettes describe some critical analyses in the manuscript in more detail. Click on the name to go to the vignette:

Key Data sets

All data are stored as dataframes with the following names. You can view additional information on each dataset by running ? before the name (e.g.; ?Comp.Game.df will show descriptions of the Comp.Game.df data)

• You can find where the package files are stored on your computer by running file.path(.libPaths(), "phillips2014rivals").
• All objects will be immediately available when you load the package. They are explicitly stored in the data_complete.RData file in the data folder in the package files.
• Raw data in comma-separated .txt files are stored in inst/extdata.

Competitive Sampling Game

Here is a description of the competitive sampling game taken verbatim from the Method section of Phillips et al. (2014):

A total of 180 students from the University of Basel par- ticipated in the study.6 They received a flat fee of CHF 7.50 (approximately $8.12 at the time) for their participation, as well as a bonus contingent on their winnings in the game. The mean bonus across both experimental conditions was CHF 1.18 (approximately $1.26) with a standard deviation of CHF 1.19. Participants completed the study in groups of four, each on a separate computer. They received no infor- mation about the choice ecology prior to beginning the task. All players began by playing three practice games without financial consequences to familiarize themselves with the experimental interface (see Appendix C for practice game parameters). They were then presented with five decision tasks. Each decision task contained two, two-outcome gambles, each with one positive and one negative outcome occurring with complementary probabilities. The gamble sets were constructed such that in certain pairs the options differed in expected value and in others they did not; like- wise, in certain pairs the options differed in range and in others they did not (see Appendix C for a full description of how gamble parameters were selected). Three different orders of each of the 12 gamble sets were created, resulting in 36 unique experimental sessions. Location of the urns on the screen was randomly determined for each decision task and on each run.

At the outset of each decision task, participants saw two options represented visually as opaque urns. They were told that each urn contained 100 virtual balls, each of which was worth a (not necessarily unique) number of points. Partici- pants were informed that they would be rewarded with one-tenth of the average value of all the balls in the urn they chose (or were allocated). Each of the participants (n = 36) assigned to the solitary condition completed one of the 36 unique experimental sessions. These participants could sample from the urns as many times as they wished before making a final choice. Having made a final choice of an urn in a decision task, they moved onto the next task. The other 144 participants played each decision task in the competition condition. At the beginning of each task, they were paired randomly with one of the other three participants. This pair- ing was done independently between tasks. Players were not told which person (of the three) they were playing against in each decision task.

Every decision task, in both the solitary and competition conditions, began with one mandatory sampling round. On every subsequent sampling round, each player indicated whether he or she wanted to sample from an urn or to make a final choice. These decisions were made privately and were only revealed to both players after both had made a sampling or choice decision. If both wanted to take a sample, they were asked to click on an urn and viewed a randomly sampled outcome from that urn. Players could see which urn the other player sampled from, but could not see the outcome the other player observed. If, after observing a sample, both players wanted another sample then another sampling round began. If one player decided to make a final choice (becoming the ‘‘chooser’’), she then selected the urn she wanted and her choice was recorded. Subsequent to the chooser’s choice, the other player (the ‘‘receiver’’) was informed that her competitor had made a choice and that he must take the remaining non- chosen urn. If both players made a choice on the same sampling round, one of two outcomes was possible: If the two players chose different urns, they each received the urn of their choice. If both players chose the same urn, the two urns were randomly assigned to the players. After final choices were made and players learned which urn they received, they were randomly paired again and the next decision task began. The random pairing was done independently of prior rounds, so a player could play the same opponent on sequential games. Participants did not receive immediate task-by-task feedback on how much money they won from their chosen urns. At the end of the session, participants were informed how much money they had earned across the five decision tasks and were paid accordingly.

Contact

• Nathaniel D. Phillips, http://ndphillips.github.io, Nathaniel.D.Phillips.is@gmail.com