Bestiary of Behavioral Economics/Activity Bias
Activity bias is when an individual, given the choice of taking action or doing nothing, chooses to take action. People want to do something rather than do nothing. Their choice to take action may result in a less-than-optimal decision being made, but they feel that taking action is what they are supposed to do. Activity bias is sometimes referred to as the active participation hypothesis.
Background
[edit | edit source]Activity bias is often used to explain errors in experimental economics. Experimental economics is the use of experimental methods in order to study questions of economics. Activity bias is present in experimental economics because some experiments may be accidentally worded in such a way or designed to encourage stupid behavior. For example, if a subject is given the choice to participate in a study or not, they might feel like they are supposed to; that they weren’t recruited for the study to do nothing.
Activity Bias in Experimental Asset Markets
[edit | edit source]One market games experiment was conducted by Vivian Lei, Charles Noussair, and Charles Plott in which there were three sections: Nospec, Twomarket, and Twomarket/Nospec. In the experiment, there were markets set up for assets with a finite number of periods. The asset then paid an allotment in each period, and that allotment was the only source of central value. This allotment is identical for each trader/subject in the experiment and the process is common knowledge to all traders/subjects. The study found that errors in decision making were resultant of activity bias.
The Asset Market Experiment
[edit | edit source]In the Nospec section, the markets are constructed similar to markets that have had bubbles and crashes. For the subject, the ability to speculate is removed and their role is either a buyer or seller so that there is no purpose to buy for the reason of resale. The only possible benefit from a purchase in this experiment is in the allotment that the asset pays out, since the unit can never be resold. In the Twomarket experiment, there is another market that operates simultaneously to the asset market. This asset market allows buying for resale and the market follows a boom and crash pattern. The directions were labeled as such to prevent bias toward inaction or action: “you are not required to participate in either of the markets if you choose not to. It may be to your advantage not to participate in either or it may be to your advantage to participate in one or both. You should decide what might be in your best interest and make your choice accordingly [1].” In the Twomarket/Nospec section, there are two markets that are exactly the same as in the Twomarket section, but the option of buying for resale in the asset market is not available.
Even with the directions not indicating a bias towards inaction or action, the subject may still feel as if they are supposed to buy and sell in the markets because they are placed into the market environment with the role of a trader [1]. Given this role, the subject may believe that buying and selling is one of the objectives of the experiment. Believing this notion, if the subject was given the choice between an unprofitable transaction and not trading, the subject will choose the unprofitable transaction. This is a speculated reason why so much trading activity occurs and the pattern of stock markets suggests such behavior occurs. In order for activity bias to be present in this study, the volume of trade must decline in the asset market in the Twomarket and Twomarket/Nospec sections relative to the standard experiments in which the asset market is the only market operating.
Results
[edit | edit source]The Results that Vivian Lei, Charles Noussair, and Charles Plott found indicated that activity bias was present because both the addition of a second market and the changes in the instructions saying that participation was optional “drastically reduced participation in asset market [1].” Activity bias in this experiment explained the large observed transaction volumes in the Nospec section; these transaction volumes were errors in decision-making. The error was believing that one had to buy and sell. The Twomarket and Twomarket/Nospec sections were designed to try to reduce the level of activity bias by stating that participation was optional.
Activity Bias in Extensive Form Games
[edit | edit source]A coordination games experiment was researched by Evren Atiker, William S. Neilson, and Michael K. Price in which they studied the effect of activity bias and focal points in a basic centipede game. "The centipede game itself consists of alternating play of binary choices, with each player making the decision to end the game or continue. Continuing the game increases the total payoff to the two players, but switches who receives the larger payoff [2]." But, if player A chose to continue and Player B did not at the next node, Player A's payoff would be lower than if A ended the game at the previous node. A single subgame perfect equilibrium (SPE) strategy combination is returned from this payoff format [2]. The effect in this equilibrium has whatever player who is the first mover ending the game at the very first node. The experiment found that activity bias was a contributing explanation for why the subjects failed to play SPE in the games.
Basis of the Experiment
[edit | edit source]The experiment used the centipede game as the basis and changed it by removing a certain confusing element and observing how it affects the game play. Changes made are small such as to make minor or no changes in strategy space, no changes in length of the game, and with one exception, to make no changes in the equilibrium path. There were four different approaches that the experiment had. The first approach held constant the joint payoff at all the decision nodes; it was the standard game but without growth in the payoff nodes. The second approach added a pair of nodes to the beginning of the game and continuing the game from this point increased both individual payoffs and joint payoffs. The third approach removed the final two nodes of the game and created a new pair of nodes. This new pair of nodes provided one player with an extremely large payoff while the other player would receive a smaller (and sometimes negative) payoff. The player who receives the larger payoff can only receive this if their opponent ends the game; this is to see the deviations from the SPE play reflect the beliefs about rationality. The fourth and final approach doesn't allow Player B to participate, the outcome of which Player A might be averse to. This approach does provide Player B with a trivial choice if Player A chooses to end the game at the first node [2].
Hypothesis of the experiment
[edit | edit source]The experiment states several hypothesis of what should happen in the game. Hypothesis three predicts that there should be no difference in the frequency of the SPE across the games for Player B in the fourth approach. The frequency of A should be grater in these games than in the games at the beginning of the experiment (standard game). This would be consistent with the explanation of activity bias and one-sided errors, but inconsistent with the prediction that players have a preference for actions that will allow their opponents to have at least one move [2].
Results
[edit | edit source]The results of the experiment found that activity bias is a contributing explanation for why the subjects failed to play the SPE in the games. Both players ended up increasing their frequency in which they play dow at the predicted node; it wasn't just Player A.
Activity Bias and Online Advertising
[edit | edit source]An experiment to test advertising effectiveness also tested activity bias and was conducted by the company Yahoo!. The research team wanted to find out if activity bias was the "source of overestimation in observational estimates of the effects of online advertising [3]." Because clicks on advertisements are easy to measure, online advertising generated a way to study the effects of advertising. There are two assumptions made: 1) if two people, A and B, looked at the exact advertisement yesterday, and only A is exposed today, then person B would be a good control to use for what would have happened to the exposed individual and 2) a person's behavior yesterday is a good indication of what they would likely do today[3]. But these assumptions fail to hold for advertising because browsing activity brings exposure to an ad campaign, making other users who saw the ad to be more active on the publisher's site during that time; different users' browsing behavior depict a large difference over time, meaning people browse a different amount of pages every day; and user's browsing behavior across the different websites appear to be positively related meaning someone who browses one website more than usual on a certain day is more likely to be viewing other sites more than usual. These three reasons combine to form the basis of activity bias; in online advertising, activity bias is the "tendency to overestimate the causal effects of advertising using online data [3]." Users browsing activity is not consistent over time and, as a result, creates a non-causal correlation between viewing the ads and purchases.
Activity bias Example
[edit | edit source]Suppose an amusement park wanted to estimate the effect of a billboard they put up along the highway would have. The company that owns the park cannot simply compare drivers who saw the ad to drivers who didn't because there are differences in the drivers. For example, drivers who did not see the billboard could be due to that they live on the other side of town. To compensate for this, the company would receive a list of drivers who travel along the highway on their morning commute. If the billboard runs for 2 weeks, then the firm would survey drivers to determine their exposure and what products they used during the 2 week period. Their study finds that the exposed drivers were more likely to come to their park with their families in the 2 week period and the week after that compared to drivers who only saw the ad once or didn't see it at all. Activity bias says the reason for the increased bias is because the drivers were more active in shopping during this period compared to the unexposed drivers. The drivers who were listed as morning commuters on that route, but didn't show up as much during this time may be because of a part-time job, they were sick or out of town, etc. It is likely the ad had a positive effect, but to reach a clear, concise conclusion that the ad was the sole reason is impossible. If that data were on hand, the amusement park company could have found there was an increase in speeding tickets "caused" by the billboard as well.
References
[edit | edit source]- ↑ a b c Vivian Lei; Charles N. Noussair; Charles R. Plott (2001), Nonspeculative Bubbles in Experimental Asset Markets: Lack of Common Knowledge or Rationality vs. Actual Irrationality (journal Economica)
- ↑ a b c d Evren Atiker; William S. Neilson; Michael K. Price (2011), Activity Bias and Focal Points in Extensive Form Games (PDF) (document)
- ↑ a b c Randall A. Lewis; Justin M. Rao; David H. Reiley (2011), Here, There, and Everywhere: Correlated Online Behaviors Can Lead to Overestimates of the Effects in Advertising (PDF)