6p no rotate
Much prep(rocessing)
How many games
## # A tibble: 1 x 3
## # Groups: numPlayers [1]
## numPlayers complete partial
## <dbl> <int> <int>
## 1 6 15 3How long full games took
## # A tibble: 1 x 7
## numPlayers games min_time `25th_time` median_time `75th_time` max_time
## <dbl> <int> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 6 15 33 49 66 74 106Many graphs
Everything here has bootstrapped 95% CIs.
Should find better curves to fit, but using quadratic to allow for some curvature.
Models
Warning many of these have less than maximal mixed effects.
## Family: bernoulli
## Links: mu = logit
## Formula: correct.num ~ block * rotate
## Data: acc_input (Number of observations: 10850)
## Samples: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
## total post-warmup samples = 4000
##
## Population-Level Effects:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## Intercept 1.65 0.07 1.52 1.78 1.00 2335 2663
## block 0.39 0.03 0.33 0.46 1.00 1686 2367
## rotatesingle -0.03 0.10 -0.22 0.16 1.00 1726 2258
## block:rotatesingle 0.04 0.05 -0.05 0.13 1.00 1602 2021
##
## Samples were drawn using sampling(NUTS). For each parameter, Bulk_ESS
## and Tail_ESS are effective sample size measures, and Rhat is the potential
## scale reduction factor on split chains (at convergence, Rhat = 1).## Family: gaussian
## Links: mu = identity; sigma = identity
## Formula: time ~ block * rotate
## Data: time_input (Number of observations: 9779)
## Samples: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
## total post-warmup samples = 4000
##
## Population-Level Effects:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## Intercept 58.12 0.71 56.70 59.49 1.00 2056 2605
## block -7.63 0.25 -8.11 -7.14 1.00 2017 2271
## rotatesingle 2.56 0.97 0.67 4.46 1.00 1800 2310
## block:rotatesingle -3.22 0.33 -3.88 -2.58 1.00 1717 2145
##
## Family Specific Parameters:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## sigma 27.87 0.20 27.48 28.26 1.00 3088 2272
##
## Samples were drawn using sampling(NUTS). For each parameter, Bulk_ESS
## and Tail_ESS are effective sample size measures, and Rhat is the potential
## scale reduction factor on split chains (at convergence, Rhat = 1).## Family: gaussian
## Links: mu = identity; sigma = identity
## Formula: words ~ block * rotate + (block | tangram) + (1 | playerId) + (1 | tangram_group) + (block | gameId)
## Data: speaker_input (Number of observations: 2170)
## Samples: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
## total post-warmup samples = 4000
##
## Group-Level Effects:
## ~gameId (Number of levels: 36)
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
## sd(Intercept) 5.21 1.29 2.61 7.74 1.01 578
## sd(block) 1.32 0.27 0.87 1.91 1.00 1189
## cor(Intercept,block) -0.77 0.16 -0.98 -0.35 1.01 510
## Tail_ESS
## sd(Intercept) 314
## sd(block) 2092
## cor(Intercept,block) 986
##
## ~playerId (Number of levels: 100)
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## sd(Intercept) 5.01 0.61 3.94 6.32 1.01 717 1273
##
## ~tangram (Number of levels: 12)
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
## sd(Intercept) 5.85 1.43 3.81 9.30 1.00 845
## sd(block) 0.87 0.24 0.50 1.44 1.00 1023
## cor(Intercept,block) -0.93 0.08 -1.00 -0.72 1.00 1525
## Tail_ESS
## sd(Intercept) 1414
## sd(block) 1560
## cor(Intercept,block) 1843
##
## ~tangram_group (Number of levels: 432)
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## sd(Intercept) 3.95 0.32 3.33 4.58 1.00 1394 1915
##
## Population-Level Effects:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## Intercept 24.82 2.34 20.23 29.48 1.00 743 1362
## block -3.52 0.56 -4.64 -2.46 1.00 964 1812
## rotatesingle 4.98 2.39 0.29 9.59 1.00 1119 1741
## block:rotatesingle -1.85 0.62 -3.05 -0.62 1.00 1175 2006
##
## Family Specific Parameters:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## sigma 8.89 0.15 8.60 9.19 1.00 2807 2463
##
## Samples were drawn using sampling(NUTS). For each parameter, Bulk_ESS
## and Tail_ESS are effective sample size measures, and Rhat is the potential
## scale reduction factor on split chains (at convergence, Rhat = 1).Do people reduce more when listeners do better?
These look pretty different from Robert’s. The word count is pretty coarse (we eliminate blatant chitchat, but didn’t subdivide lines with some referential language). More is said when people get things wrong than right, but it’s not clear if it’s reduction vs how hard it was or where they were starting from.
For bonus fun, look at all the rotate data.
NLP prep
Content analyses
Of words the speaker says in the last round, when were they said by the speaker in earlier rounds for the same tangram?
(TODO vector analysis)