TODO notes:

Study 1 notes

2x2 study of BoS vs PD crossed with chat or no-chat. All did 40 trials. All did a 3 minute “turing-test” free chat at the start. Run Jan 7. Target was 20 pairs in each condition.

Pre-reg

According to https://osf.io/8fnze, the analyses I’m going to do:

This is exploratory research, here are some things we plan to explore. role of language on score. For each type of game, do games with language have higher overall scores ? Is this related to if language was actually used? Quantity and type of language over rounds. How much language is produced each round in PD v BoS? Does it decline in later rounds? What type of language is used? Is it just identifying what to select or is there meta-strategy/negotiation? Are pairs converging to predictable strategies?

Pre-process

Game length

Estimated 15 mins total including intro/exit/waiting turing etc

How long did people spend playing? (Not counting intro chat round)

## # A tibble: 4 x 3
## # Groups:   game_cond [2]
##   game_cond chat_cond     n
##   <chr>     <chr>     <int>
## 1 BoS       chat         19
## 2 BoS       nochat       16
## 3 PD        chat         17
## 4 PD        nochat       17

A respectable rate of game completion.

Bonuses

Note that bonuses are not directly comparable between BoS & PD! But it looks like chat helps in BoS and not (much?) in PD.

Pre-chat

People talked some. I haven’t looked at the transcripts at all.

During chat

This chat is raw. There are outliers that have been chopped off. People talk more during BoS than PD.

Currently open questions – is that game-related chat, or is it just that it’s a friendlier game? Does the chatting help?

What are people saying?

## # A tibble: 1 x 1
##       s
##   <int>
## 1  2093

Most utterances are single word utterances, we’ll consider these and multi word utterances separately.

##      WORD FREQ
## 1   green   90
## 2      ok   84
## 3    blue   83
## 4    pink   56
## 5  purple   45
## 6     red   38
## 7  yellow   35
## 8  orange   26
## 9   brown   24
## 10      g   22
## 11   grey   18
## 12   okay   16
## 13      u   15
## 14      b   12
## 15      y   11
## 16      I    9
## 17      p    9
## 18      r    9
## 19  white    9
## 20   yeah    8
## [1] 433

There are 743 singletons. Of these 100 are ok/okay, at least 433 are color words, more once you count rarities and misspellings. There’s also a bunch of singleton letters – these seem to be single letter color abbreviations?

From skimming transcripts, looks like multi’s occur earlier in general. But we’re confounded by pleasantries at random times (talking about where they live, etc).

game_cond repNum text
BoS 0 are we splitting or out to make money
BoS 0 id like ot be fair
BoS 0 trying to figure out how to do that tho
BoS 0 i think we need a few rounds to figure it out
BoS 0 there must be some rules
BoS 0 ok green together
BoS 2 i pink you blue
BoS 2 or wait
BoS 3 ok just go by the chart
BoS 3 we need to select the same
BoS 4 you can have it take yellow
BoS 4 yellow ill give the next to you
BoS 6 take it
BoS 7 awesome team work
BoS 7 hell yeaah
BoS 7 left one
BoS 7 the purplish
BoS 11 awesome d
BoS 16 i think its different for us
BoS 16 the positions of the colors
BoS 16 the green
BoS 23 i love the teamwork
BoS 23 hahaah yess
BoS 23 where are you from
BoS 23 im curious
BoS 23 germany and you
BoS 23 ich bin von polen
BoS 23 noice d
BoS 23 lets go green
BoS 35 go pink
BoS 39 oh woow last round
BoS 39 you can have it
BoS 5 one for me one for you
BoS 5 i was going one left one right
BoS 5 do blue
BoS 8 blue one xd
BoS 8 the other
BoS 9 one for me now
BoS 9 now the blue one
BoS 10 now pink

Does chat help?

To look at if chatting helps – we can compare forced no-chat to chat (see bonuses above). We can also compare used chat (0/1 or amount) within those that could chat.

Options for dependent variable are bonus (although there’s some noise there) or option chosen.

Looks like chatting helps for BoS, doesn’t get used much for PD.

Choices

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

In PD: P1 prefers BA > AA > BB > AB and P2 prefers AB > AA > BB > BA. (BB is Nash equilibrium, AA is Pareto dominant.)

Not really sure how to visualize this efficiently?

Being in the chat condition v not

Actually using the chat that round

Is the other player a human?

## # A tibble: 4 x 4
## # Groups:   game_cond, chat_cond [4]
##   game_cond chat_cond    no   yes
##   <chr>     <chr>     <int> <int>
## 1 BoS       chat          4    34
## 2 BoS       nochat        6    26
## 3 PD        chat          5    29
## 4 PD        nochat        4    30

Only 10-20% of players think the other person is a bot!