Pre-process

Read data

Summary of experiment

In the expt reported here pairs of participants played 40 rounds of a game-theory type game. At the start, each pair had 3 minutes of free chat, and then played the game. We recruited for 20 games in chat and 20 games in no-chat conditions.

## # A tibble: 2 x 3
## # Groups:   game_cond [1]
##   game_cond chat_cond     n
##   <chr>     <chr>     <int>
## 1 jun24     chat         14
## 2 jun24     nochat       20

Fewer chat games than non-chat games – not sure why the differential loss – maybe due to order of running??

Pre-reg’d analysis plan

This is exploratory research, here are comparisons we plan to make (i.e. with graphs).

  • Do games with the option to chat get better scores than games without (total, and subdivided by trial type)?

  • Which quadrant (combination of A,B responses) do players land in as a function of chat/nochat and trial condition? (Do games with chat choose the off-diagonal option in “sacrifice” PD trials?)

  • How much language do players use as a function of trial number and type?

  • (More qualitative) What heuristics/strategies do players use? (gleaned from post-experiment question and chat logs) How successful are they and which ones require how much language to coordinate?

  • Are there signs of conventionalizing on ways of describing trials / strategies beyond names for boxes?

other stuff to explore

  • need to do intent to treat analysis
  • look at what players write as their strategy
  • can we tell what heuristics are being used (by trial, as function of alternating, as function of score so far?)
  • correlation w/i game between outcomes on different trial types

Overall outcomes

A coarse first pass is to look at how much bonus people earned, as bonus is propotional to points.

By trial type:

Aggregate:

Clear aggregate difference based on condition.

Pre-chat

We hope that players use the pre-chat to communicate with each other and convince themselves that the other person is a human. All games are the same at this point, so we don’t expect differential amounts of talk.

During chat

Chatting seems to increase slightly across the games, but many groups even in the chat condition don’t use the chat.

Are they human?

## # A tibble: 2 x 5
## # Groups:   game_cond, chat_cond [2]
##   game_cond chat_cond    no   yes   pct
##   <chr>     <chr>     <int> <int> <dbl>
## 1 jun24     chat          3    25 0.893
## 2 jun24     nochat        8    32 0.8

Yay, they mostly think they’re human!!

Chat and payoff

We look at how many rounds had non-zero chat (of those with the option to chat) compared to mean payoff.

Using the chat more of the time, and using more words are both associated with higher payoffs.

We can break this down by trial type.

Chat helps a lot with BoS and spikeBos, maybe helps a little with hardPD, does not help with easyPD.

We can go even more granular and look at whether words on a trial is associated with more bonus * on that trial *.

There are some real outlier rounds where people say a lot, but mostly they say not that much even when they do say things. Talking is better than not talking for most trial types, but there’s not much of a dose-response pattern.

Chat and choices

Rather than collapsing over outcome (which has some randomness), we can look at per-round, which quadrant in the game people chose.

A number of people in the “chat” condition didn’t use chat, so we sometimes 3 way split for used chat / didn’t use chat / couldn’t chat (in contrast to just spliting by intent to treat).

normal BoS

In BoS: P1 prefers AA to BB, P2 prefers BB to AA. AB and BA are bad for both.

Near chance if you don’t talk, far above chance if you do.

spike BoS

In spike BoS: P1 prefers AA to BB, P2 prefers BB to AA. AB and BA are bad for both. (same distribution for spikeBoS, just more skewed).

Actually talking is conducive to good outcomes.

easyPD

In easyPD: P1 prefers BA > AA > BB > AB and P2 prefers AB > AA > BB > BA. AA is welfare maximizing.

Everyone is likely to choose the good outcome, but those who talk are more reliable at it (especially early on). No real consistency in what is chosen otherwise.

hardPD

In hardPD: P1 prefers BA > AA > BB > AB and P2 prefers AB > AA > BB > BA. BA and AB are welfare maximizing.

When chat isn’t used, AA increases over the course of the game. When chat is used, AB/BA increases. These are mostly trading off with each other, the obviously bad BB option isn’t selected much ever.

When do people talk?

Seems to be substantial group to group variability in if they talk never/ sometimes/ always. Isn’t fully driven by trial type.

How do we increase the rates of talkers!!??

Does doing well in one type mean doing well in another??

For each game, look at fraction of each trial type where there did which quadrant and correlate??

What do they say strategy is?

chat_cond strategy
chat be nice and not greedy
chat Try to keep it balanced
chat Choose the best option aviable
chat me and my partner worked together very well, so we both got the same pay at the end.
chat One person was getting the most out of every other box, unless both boxes were quite equal
chat We gave each other fair chances to get the most rewards.
chat maximise paymeny for the both of us
chat I wanted each of us to get an equal number of points
chat Initially I wanted to make sure everyone got the same. It was the partner’s suggestion as well: even if we got less than maximum value, everyone gets something. But the partner’s strategy changed, leaving me with less at the start. So I did try to compensate so that everyone got more or less the same.
chat I initially wanted to talk out every choice with my partner, because we could have maximized our bonus, but they didn’t seem to want to, so we tried to just nonverbally alternate giving each other large amounts of points.
chat I was trying to click the box that benefited us both.
chat i started trying to get the money for me but we got more when we coordinated
chat I accidentally clicked wrong on the first box and my partner thought i was against him, but i wasnt. but after we developed a strategy
chat I tried to balance the bonus between me and my partner.
chat Maximize the amount i could earn with the occasional switch up.
chat To take advantage with the best possibility to earn more money
chat I tried to take turns with my partner to take the higher prize.
chat To do most equal, and make up the next rounds to make it even
chat I tried to cooperate with my partner so that the distribution would be equitable and each round we would each take a little more. Since he didn’t answer me in the chat, I followed his strategy, which seemed to be to end both with the same money and always give the most equal answer.
chat Maximize gain for both
chat try to keep both scores balanced
chat Kept an open communication and both shared the wealth of each round.
chat If the points were higher on either side we take turns in earning the highest points. If the points were equal we go for the same box.
chat I just tried to split the bonus rewards has equally as I could, while talking to the other person if it was okay to go for the choices.
chat we always talked and all and it was extremely fun
chat Me and my partner always selected the option that benefited both unless there was a gap between bonus payments. In that case we tried to close the gap.
chat It was to get the maximum possible bonus for both of us and to make sure that we both got almost the same amount.
chat do be fair
nochat Choose the option that paid us both, favoring my opponent to be sure they’d play along
nochat We tried to pick the ones that would let us both earn
nochat guessing
nochat I was trying to make decisions which will in the end give us equal reward
nochat I tried to choose what would be fair for both of us, but it was sometimes difficult.
nochat Try to always give the same pay for both
nochat If I won more than my partnet, on the next turn I chose the best answer for him or her
nochat I opted for the highest fairest distribution when I could, and switched between the choices that favored them or me when I couldn’t to match a few
nochat We just went with the best option that balanced us both out in the end on the earnings.
nochat I tried to choose options which would let my partner to understand I was working towards a common goal
nochat No strategy
nochat pick the opposite of whatever I thought it was
nochat I didn’t have one
nochat Be coordinated with my partner and select the equal rewards.
nochat Varying between maximizing my own profit and following my partners lead on ocasion
nochat Try to see what was the best outcome for everyone
nochat See how many points my partner have in each box, and then choose the best option
nochat I tried to make him win more or both equally
nochat I was trying to maximise my payoff versus my partner.
nochat Pick the best one for both of us
nochat Choose the option that provides the best outcome for me without my partner earning more than me.
nochat When the compensation was equal, such as 4-4 , I chose that. When there was 0-0 and more money for my partner, I gave it to him.
nochat Initially try and be fair. When I found my partner to be greedy (and ALWAYS choose the highest option for themselves) I thought it best to play along so as to make sure I won something.
nochat In situations where the difference between points was big i tried to get the most of it, but not always, to confuse my partner a bit. I was trying to get 50/50 split when i felt it was the best for both parties
nochat the strategy was to benefit as much, and this meant that I had to work together with the other player.
nochat We cooperated as best we could to maximize benefit for both of us
nochat Take a lower return if this was mutually beneficial in order to build trust
nochat I wanted to choose option that will be equally beneficial to me and my partner
nochat Choose the square with the numbers in my favour.
nochat Try to have us winning roughly the equal amounts. So if one of us was ahead, try to offer the other the most points. Or sometimes take it in turn to get the most points.
nochat I try to get the best option for both of us. But also reacted to his/her decision if he/she wants to get more than me.
nochat Select the box that would gain the maximum for both players
nochat Whenever an option came up to mutually win evenly, I selected that option and was proud to see that my partner did the same, even though it would have mathematically been in our favor to "betray" each other strategically when the sum of the numbers is the highest, if we were to always err on the side of the person with the lowest total earnings. My partner always chose the mutually beneficial option, and I always did the same, especially since that kept the trust up. On the lopsided ones, I think my partner initially wanted to alternate between what benefits me and what benefits him, but I was going based on who was earning the least in that moment. I believe we realized that and switched to the other person’s strategy, which actually made us lose a few rounds since we both did so at the same time. Eventually it seems we landed on giving the extra earnings to the person with the lower running total, and through this strategy and luck we ended up getting the same exact amount in the end.
nochat I didn’t want to seem like I was cheating the other person out of money. I tried to have a pattern that the other person would pick up on so we could get more money.
nochat I tried to predict my buddy’s decision, sometimes worked out but sometimes it did not
nochat I tried to even the rewards
nochat For ones where there is an option for both of us to get even amounts, we chose that option. For options where one would get the higher, I usually chose the option that would be better for the other person so we both would get something
nochat I didn’t have any specific strategy
nochat nothing my partner was awful
nochat Picking a color that would benefit me and them either way