-
Notifications
You must be signed in to change notification settings - Fork 0
/
football_ms_SI.Rmd
429 lines (316 loc) · 28.6 KB
/
football_ms_SI.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
---
title: "Supplementary Material for 'Cultural evolution of football tactics: Strategic social learning in managers' choice of formation'"
author: Alex Mesoudi ^[Human Behaviour and Cultural Evolution Group, Department of Biosciences, College of Life and Environmental Sciences, University of Exeter, Penryn, Cornwall TR10 9FE, United Kingdom; [email protected]]
date: '`r format(Sys.time(), "%d %B, %Y")`'
fontsize: 11pt
output:
bookdown::pdf_document2:
fig_caption: yes
fig_height: 4
fig_width: 4
number_sections: yes
toc: true
header-includes:
- \usepackage{setspace}
- \doublespacing
#- \singlespacing
- \usepackage{lineno}
#- \linenumbers
- \def\linenumberfont{\normalfont\small\sffamily}
#- \def\linenumberfont{\normalfont\tiny\sffamily}
- \newcommand{\beginsupplement}{\setcounter{table}{0} \renewcommand{\thetable}{S\arabic{table}} \setcounter{figure}{0} \renewcommand{\thefigure}{S\arabic{figure}}}
csl: nature.csl
bibliography: football.bib
link-citations: yes
---
\beginsupplement
\newpage
# Model specifications
The following code gives the models as specified in the rethinking package [@mcelreath_rethinking:_2019] in R, using non-centred parameterisation.
## Null model
```{r null_code, eval = F}
fourtwothreeone ~ dbinom(1, p)
logit(p) <- a_homeaway[homeaway] + b_team_strength * team_strength +
a_manager[manager_id] * sigma_manager + a_division[division_id] *
sigma_division
a_homeaway[homeaway] ~ dnorm(0, 1)
b_team_strength ~ dnorm(0, 1)
a_manager[manager_id] ~ dnorm(0, 1)
a_division[division_id] ~ dnorm(0, 1)
sigma_manager ~ dexp(1)
sigma_division ~ dexp(1)
```
## Personal model
```{r pers_model_code, eval = F}
fourtwothreeone ~ dbinom(1, p)
logit(p) <- a_homeaway[homeaway] + b_team_strength * team_strength +
a_manager[manager_id, 1] + a_manager[manager_id, 2] * pers_use +
a_division[division_id, 1] + a_division[division_id, 2] *
pers_use + b_pers_use * pers_use + b_pers_win * pers_win +
b_pers_useXpers_win * pers_useXpers_win + b_pers_useXpers_win_all *
pers_useXpers_win_all
a_homeaway[homeaway] ~ dnorm(0, 1)
b_team_strength ~ dnorm(0, 1)
transpars > matrix[manager_id, 2]:a_manager <- compose_noncentered(sigma_manager,
L_Rho_manager, z_manager)
transpars > matrix[division_id, 2]:a_division <- compose_noncentered(sigma_division,
L_Rho_division, z_division)
matrix[2, manager_id]:z_manager ~ normal(0, 1)
matrix[2, division_id]:z_division ~ normal(0, 1)
b_pers_use ~ dnorm(0, 1)
b_pers_win ~ dnorm(0, 1)
b_pers_useXpers_win ~ dnorm(0, 1)
b_pers_useXpers_win_all ~ dnorm(0, 1)
vector[2]:sigma_manager ~ dexp(1)
cholesky_factor_corr[2]:L_Rho_manager ~ lkj_corr_cholesky(2)
vector[2]:sigma_division ~ dexp(1)
cholesky_factor_corr[2]:L_Rho_division ~ lkj_corr_cholesky(2)
```
## Population model
```{r pop_model_code, eval = F}
fourtwothreeone ~ dbinom(1, p)
logit(p) <- a_homeaway[homeaway] + b_team_strength * team_strength +
a_manager[manager_id, 1] + a_manager[manager_id, 2] * pop_use +
a_division[division_id, 1] + a_division[division_id, 2] *
pop_use + b_pop_use * pop_use + b_pop_win * pop_win + b_pop_useXpop_win *
pop_useXpop_win + b_pop_useXpers_win_all * pop_useXpers_win_all
a_homeaway[homeaway] ~ dnorm(0, 1)
b_team_strength ~ dnorm(0, 1)
transpars > matrix[manager_id, 2]:a_manager <- compose_noncentered(sigma_manager,
L_Rho_manager, z_manager)
transpars > matrix[division_id, 2]:a_division <- compose_noncentered(sigma_division,
L_Rho_division, z_division)
matrix[2, manager_id]:z_manager ~ normal(0, 1)
matrix[2, division_id]:z_division ~ normal(0, 1)
b_pop_use ~ dnorm(0, 1)
b_pop_win ~ dnorm(0, 1)
b_pop_useXpop_win ~ dnorm(0, 1)
b_pop_useXpers_win_all ~ dnorm(0, 1)
vector[2]:sigma_manager ~ dexp(1)
cholesky_factor_corr[2]:L_Rho_manager ~ lkj_corr_cholesky(2)
vector[2]:sigma_division ~ dexp(1)
cholesky_factor_corr[2]:L_Rho_division ~ lkj_corr_cholesky(2)
```
## Full model
```{r full_model_code, eval = F}
fourtwothreeone ~ dbinom(1, p)
logit(p) <- a_homeaway[homeaway] + b_team_strength * team_strength +
a_manager[manager_id, 1] + a_manager[manager_id, 2] * pers_use +
a_manager[manager_id, 3] * pop_use + a_division[division_id,
1] + a_division[division_id, 2] * pers_use + a_division[division_id,
3] * pop_use + b_vars
b_vars <- b_pers_use * pers_use + b_pers_win * pers_win + b_pers_useXpers_win *
pers_useXpers_win + b_pers_useXpers_win_all * pers_useXpers_win_all +
b_pop_use * pop_use + b_pop_win * pop_win + b_pop_useXpop_win *
pop_useXpop_win + b_pop_useXpers_win_all * pop_useXpers_win_all
a_homeaway[homeaway] ~ dnorm(0, 1)
b_team_strength ~ dnorm(0, 1)
transpars > matrix[manager_id, 3]:a_manager <- compose_noncentered(sigma_manager,
L_Rho_manager, z_manager)
transpars > matrix[division_id, 3]:a_division <- compose_noncentered(sigma_division,
L_Rho_division, z_division)
matrix[3, manager_id]:z_manager ~ normal(0, 1)
matrix[3, division_id]:z_division ~ normal(0, 1)
b_pers_use ~ dnorm(0, 1)
b_pers_win ~ dnorm(0, 1)
b_pers_useXpers_win ~ dnorm(0, 1)
b_pers_useXpers_win_all ~ dnorm(0, 1)
b_pop_use ~ dnorm(0, 1)
b_pop_win ~ dnorm(0, 1)
b_pop_useXpop_win ~ dnorm(0, 1)
b_pop_useXpers_win_all ~ dnorm(0, 1)
vector[3]:sigma_manager ~ dexp(1)
cholesky_factor_corr[3]:L_Rho_manager ~ lkj_corr_cholesky(2)
vector[3]:sigma_division ~ dexp(1)
cholesky_factor_corr[3]:L_Rho_division ~ lkj_corr_cholesky(2)
```
## Manager model
```{r manager_model_code, eval = F}
win_rate ~ dnorm(mu, sigma)
mu <- a + b1 * info_ratio + b2 * info_ratio_sq
a ~ normal(0, 1.5)
b1 ~ normal(0, 0.2)
b2 ~ normal(-0.1, 0.2)
sigma ~ dexp(1)
```
\newpage
# Results with different predictor time windows
The analyses in the main text use a time window of $X=30$ days to calculate all use and win rate predictors. Here I present results with different time windows: $X=20$, $X=40$ and $X=60$ (NB these choices, both what to include in the main paper and alternatives to consider here, were all preregistered).
## Hypothesis H1: combination of population and personal information use
Tables S1-S3 show model comparisons for X=20, X=40 and X=60 day windows respectively, to compare with Table 1 in the main paper that uses X=30. Across all time windows, the full model is best supported, receiving all model weight, and the personal model was next best supported.
Tables S4-S6 show the full models for X=20, X=40 and X=60 day windows, to compare with Table 2 in the main paper that uses X=30. Apart from inevitable differences in coefficient values, the models all have the same patterns of results: for each time window, the same predictors either cross zero or do not cross zero.
```{r tableS1, echo = F, message = F}
mcomparison <- as.matrix(readRDS(file="modeloutputs/mcomparisonX20.rds"))
row.names(mcomparison)[row.names(mcomparison) == "m1"] <- "Null model"
row.names(mcomparison)[row.names(mcomparison) == "m2"] <- "Personal model"
row.names(mcomparison)[row.names(mcomparison) == "m3"] <- "Population model"
row.names(mcomparison)[row.names(mcomparison) == "m4"] <- "Full model"
knitr::kable(mcomparison, caption="Model comparison to test hypothesis H1, with X=20 day time window.", digits = 2, booktabs = T, linesep="")
```
```{r tableS2, echo = F, message = F}
mcomparison <- as.matrix(readRDS(file="modeloutputs/mcomparisonX40.rds"))
row.names(mcomparison)[row.names(mcomparison) == "m1"] <- "Null model"
row.names(mcomparison)[row.names(mcomparison) == "m2"] <- "Personal model"
row.names(mcomparison)[row.names(mcomparison) == "m3"] <- "Population model"
row.names(mcomparison)[row.names(mcomparison) == "m4"] <- "Full model"
knitr::kable(mcomparison, caption="Model comparison to test hypothesis H1, with X=40 day time window.", digits = 2, booktabs = T, linesep="")
```
```{r tableS3, echo = F, message = F}
mcomparison <- as.matrix(readRDS(file="modeloutputs/mcomparisonX60.rds"))
row.names(mcomparison)[row.names(mcomparison) == "m1"] <- "Null model"
row.names(mcomparison)[row.names(mcomparison) == "m2"] <- "Personal model"
row.names(mcomparison)[row.names(mcomparison) == "m3"] <- "Population model"
row.names(mcomparison)[row.names(mcomparison) == "m4"] <- "Full model"
knitr::kable(mcomparison, caption="Model comparison to test hypothesis H1, with X=60 day time window.", digits = 2, booktabs = T, linesep="")
```
```{r tableS4, echo = F, message = F}
precis_m4 <- as.matrix(readRDS(file="modeloutputs/precis_m4X20.rds"))
# add rows, sub-headings for fixed and varying effects
precis_m4 <- rbind(fixed_effects = rep(NA,6), precis_m4)
precis_m4 <- rbind(precis_m4[1:12,], rep(NA,6), varying_effects = rep(NA,6), precis_m4[13:18,])
row.names(precis_m4) <- c("Fixed effects:", "Home/away: Home", "Home/away: Away", "Team strength", "Personal 4231 use", "Personal 4231 win rate", "Personal 4231 use * personal 4231 win rate", "Personal 4231 use * personal win rate", "Population 4231 use", "Population 4231 win rate", "Population 4231 use * population 4231 win rate", "Population 4231 use * personal win rate", " ", "Varying effects:", "Manager", "Manager * personal 4231 use", "Manager * population 4231 win rate", "Division", "Division * personal 4231 use", "Division * population 4231 win rate")
options(knitr.kable.NA = '')
knitr::kable(precis_m4[,1:4], caption="Parameter estimates for the full model, with X=20 day time window. Home/away is an indicator trait with separate estimates for formations used home and away. Varying effects show the standard deviations of the varying intercepts and slopes. See SI for full model specification and priors.", digits = 2, booktabs = T, linesep="")
```
```{r tableS5, echo = F, message = F}
precis_m4 <- as.matrix(readRDS(file="modeloutputs/precis_m4X40.rds"))
# add rows, sub-headings for fixed and varying effects
precis_m4 <- rbind(fixed_effects = rep(NA,6), precis_m4)
precis_m4 <- rbind(precis_m4[1:12,], rep(NA,6), varying_effects = rep(NA,6), precis_m4[13:18,])
row.names(precis_m4) <- c("Fixed effects:", "Home/away: Home", "Home/away: Away", "Team strength", "Personal 4231 use", "Personal 4231 win rate", "Personal 4231 use * personal 4231 win rate", "Personal 4231 use * personal win rate", "Population 4231 use", "Population 4231 win rate", "Population 4231 use * population 4231 win rate", "Population 4231 use * personal win rate", " ", "Varying effects:", "Manager", "Manager * personal 4231 use", "Manager * population 4231 win rate", "Division", "Division * personal 4231 use", "Division * population 4231 win rate")
options(knitr.kable.NA = '')
knitr::kable(precis_m4[,1:4], caption="Parameter estimates for the full model, with X=40 day time window. Home/away is an indicator trait with separate estimates for formations used home and away. Varying effects show the standard deviations of the varying intercepts and slopes. See SI for full model specification and priors.", digits = 2, booktabs = T, linesep="")
```
```{r tableS6, echo = F, message = F}
precis_m4 <- as.matrix(readRDS(file="modeloutputs/precis_m4X60.rds"))
# add rows, sub-headings for fixed and varying effects
precis_m4 <- rbind(fixed_effects = rep(NA,6), precis_m4)
precis_m4 <- rbind(precis_m4[1:12,], rep(NA,6), varying_effects = rep(NA,6), precis_m4[13:18,])
row.names(precis_m4) <- c("Fixed effects:", "Home/away: Home", "Home/away: Away", "Team strength", "Personal 4231 use", "Personal 4231 win rate", "Personal 4231 use * personal 4231 win rate", "Personal 4231 use * personal win rate", "Population 4231 use", "Population 4231 win rate", "Population 4231 use * population 4231 win rate", "Population 4231 use * personal win rate", " ", "Varying effects:", "Manager", "Manager * personal 4231 use", "Manager * population 4231 win rate", "Division", "Division * personal 4231 use", "Division * population 4231 win rate")
options(knitr.kable.NA = '')
knitr::kable(precis_m4[,1:4], caption="Parameter estimates for the full model, with X=60 day time window. Home/away is an indicator trait with separate estimates for formations used home and away. Varying effects show the standard deviations of the varying intercepts and slopes. See SI for full model specification and priors.", digits = 2, booktabs = T, linesep="")
```
## Hypothesis H2: ratio of population to personal 4231 use
For $X=20$, the mean population:personal use ratio was 0.92, 89%CI[0.40, 1.63]. For $X=40$ it was 0.68, 89%CI[0.16, 1.41]. For $X=60$ it was 0.54, 89%CI[0.05, 1.25]. For all time windows, therefore, the ratio was not only not greater than one as was predicted, but the CIs are in all cases so wide that they encompass one. There is, however, a trend towards lower ratios with longer time windows, indicating that assuming longer time windows leads to greater reliance on personal information over population information.
## Hypothesis H3: variation across managers and divisions
Tables S7-S9 show that for $X=20$, $X=40$ and $X=60$, as for $X=30$, there is more varation in 4231 personal and population use across managers than randomised data, and more variation across divisions in personal 4231 use, but not across divisions in population 4231 use.
```{r tableS7, echo = F, message = F}
precis_sigmadiffs <- as.matrix(readRDS(file="modeloutputs/precis_sigmadiffsX20.rds"))
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pers_manager"] <- "Manager * personal 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pop_manager"] <- "Manager * population 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pers_division"] <- "Division * personal 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pop_division"] <- "Division * population 4231 use"
knitr::kable(precis_sigmadiffs, caption="Tests of the differences between varying effects from the real data and varying effects from randomised data, to test hypothesis H3 with X=20 day time window. Values shown are real minus randomised standard deviations.", digits = 2, booktabs = T, linesep="")
```
```{r tableS8, echo = F, message = F}
precis_sigmadiffs <- as.matrix(readRDS(file="modeloutputs/precis_sigmadiffsX40.rds"))
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pers_manager"] <- "Manager * personal 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pop_manager"] <- "Manager * population 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pers_division"] <- "Division * personal 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pop_division"] <- "Division * population 4231 use"
knitr::kable(precis_sigmadiffs, caption="Tests of the differences between varying effects from the real data and varying effects from randomised data, to test hypothesis H3 with X=40 day time window. Values shown are real minus randomised standard deviations.", digits = 2, booktabs = T, linesep="")
```
```{r tableS9, echo = F, message = F}
precis_sigmadiffs <- as.matrix(readRDS(file="modeloutputs/precis_sigmadiffsX60.rds"))
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pers_manager"] <- "Manager * personal 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pop_manager"] <- "Manager * population 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pers_division"] <- "Division * personal 4231 use"
row.names(precis_sigmadiffs)[row.names(precis_sigmadiffs) == "diff_pop_division"] <- "Division * population 4231 use"
knitr::kable(precis_sigmadiffs, caption="Tests of the differences between varying effects from the real data and varying effects from randomised data, to test hypothesis H3 with X=60 day time window. Values shown are real minus randomised standard deviations.", digits = 2, booktabs = T, linesep="")
```
## Hypothesis H4: population to personal use ratio and win rate
Tables S10-S12 show, as for $X=30$, that there is no quadratic or linear relationship between manager win rate and population:personal use ratio for $X=20$, $X=40$ or $X=60$.
```{r tableS10, echo = F, message = F}
precis_manager <- as.matrix(readRDS(file="modeloutputs/precis_managerX20.rds"))
knitr::kable(precis_manager, caption="Model estimates for the quadratic regression model with manager as unit of analysis, to test hypothesis H4 with X=20 day time window. Parameter a is the intercept, b1 is the linear coefficient and b2 the quadratic coefficient. Win rate is modelled as normally distributed with standard deviation sigma. See SI for priors.", digits = 2, booktabs = T, linesep="")
```
```{r tableS11, echo = F, message = F}
precis_manager <- as.matrix(readRDS(file="modeloutputs/precis_managerX40.rds"))
knitr::kable(precis_manager, caption="Model estimates for the quadratic regression model with manager as unit of analysis, to test hypothesis H4 with X=40 day time window. Parameter a is the intercept, b1 is the linear coefficient and b2 the quadratic coefficient. Win rate is modelled as normally distributed with standard deviation sigma. See SI for priors.", digits = 2, booktabs = T, linesep="")
```
```{r tableS12, echo = F, message = F}
precis_manager <- as.matrix(readRDS(file="modeloutputs/precis_managerX60.rds"))
knitr::kable(precis_manager, caption="Model estimates for the quadratic regression model with manager as unit of analysis, to test hypothesis H4 with X=60 day time window. Parameter a is the intercept, b1 is the linear coefficient and b2 the quadratic coefficient. Win rate is modelled as normally distributed with standard deviation sigma. See SI for priors.", digits = 2, booktabs = T, linesep="")
```
\newpage
# Revised models excluding own team formations from population predictors
The following results are from the revised models which, for the population predictors, exclude formations used by the same team during the $X$-day window. That is, for team $i$, the population predictors are calculated from all formations used in the $X$-day window except those used by team $i$. This ensures a stricter separation between personal and population predictors, as now team $i$'s formations are only used in the personal predictors, not the population predictors. This change had negligible effects on the results, and all conclusions for all hypotheses remain qualitatively identical to the original findings derived from the original preregistered analyses presented in the main paper.
## Hypothesis H1: combination of population and personal information use
Table S13 shows almost identical model comparison for the revised models excluding own team from population predictors, compared to the original model given in the main text. Table S14 also shows very similar estimates for the full revised model compared to the original.
```{r tableS13, echo = F, message = F}
mcomparisonR1 <- as.matrix(readRDS(file="modeloutputs/mcomparisonR1.rds"))
row.names(mcomparisonR1)[row.names(mcomparisonR1) == "m1R"] <- "Null model"
row.names(mcomparisonR1)[row.names(mcomparisonR1) == "m2R"] <- "Personal model"
row.names(mcomparisonR1)[row.names(mcomparisonR1) == "m3R"] <- "Population model"
row.names(mcomparisonR1)[row.names(mcomparisonR1) == "m4R"] <- "Full model"
knitr::kable(mcomparisonR1, caption="Model comparison to test hypothesis H1, with revised models excluding own team from the population predictors.", digits = 2, booktabs = T, linesep="")
```
```{r tableS14, echo = F, message = F}
precis_m4R1 <- as.matrix(readRDS(file="modeloutputs/precis_m4R1.rds"))
# add rows, sub-headings for fixed and varying effects
precis_m4R1 <- rbind(fixed_effects = rep(NA,6), precis_m4R1)
precis_m4R1 <- rbind(precis_m4R1[1:12,], rep(NA,6), varying_effects = rep(NA,6), precis_m4R1[13:18,])
row.names(precis_m4R1) <- c("Fixed effects:", "Home/away: Home", "Home/away: Away", "Team strength", "Personal 4231 use", "Personal 4231 win rate", "Personal 4231 use * personal 4231 win rate", "Personal 4231 use * personal win rate", "Population 4231 use", "Population 4231 win rate", "Population 4231 use * population 4231 win rate", "Population 4231 use * personal win rate", " ", "Varying effects:", "Manager", "Manager * personal 4231 use", "Manager * population 4231 win rate", "Division", "Division * personal 4231 use", "Division * population 4231 win rate")
options(knitr.kable.NA = '')
knitr::kable(precis_m4R1[,1:4], caption="Parameter estimates for the full revised model excluding own team from the population predictors. Home/away is an indicator trait with separate estimates for formations used home and away. Varying effects show the standard deviations of the varying intercepts and slopes. See SI for full model specification and priors.", digits = 2, booktabs = T, linesep="")
```
## Hypothesis H2: ratio of population to personal 4231 use
For the revised model excluding own team from population predictors, the mean population:personal use ratio was 0.79, 89%CI[0.21, 1.48]. Similar to the original model, the mean ratio is less than one, and the CIs are so wide that they encompass one.
## Hypothesis H3: variation across managers and divisions
Table S15 shows that for the revised model excluding own team from population predictors, there is more varation in 4231 personal and population use across managers than randomised data, and more variation across divisions in personal 4231 use, but not across divisions in population 4231 use. This is the same as the original model.
```{r tableS15, echo = F, message = F}
precis_sigmadiffsR1 <- as.matrix(readRDS(file="modeloutputs/precis_sigmadiffsR1.rds"))
row.names(precis_sigmadiffsR1)[row.names(precis_sigmadiffsR1) == "diff_pers_manager"] <- "Manager * personal 4231 use"
row.names(precis_sigmadiffsR1)[row.names(precis_sigmadiffsR1) == "diff_pop_manager"] <- "Manager * population 4231 use"
row.names(precis_sigmadiffsR1)[row.names(precis_sigmadiffsR1) == "diff_pers_division"] <- "Division * personal 4231 use"
row.names(precis_sigmadiffsR1)[row.names(precis_sigmadiffsR1) == "diff_pop_division"] <- "Division * population 4231 use"
knitr::kable(precis_sigmadiffsR1, caption="Tests of the differences between varying effects from the real data and varying effects from randomised data, to test hypothesis H3 with the revised model excluding own team from the population predictors. Values shown are real minus randomised standard deviations.", digits = 2, booktabs = T, linesep="")
```
## Hypothesis H4: population to personal use ratio and win rate
Table S16 shows that there is no quadratic or linear relationship between manager win rate and population:personal use ratio for the revised model, just as for the original model.
```{r tableS16, echo = F, message = F}
precis_managerR1 <- as.matrix(readRDS(file="modeloutputs/precis_managerR1.rds"))
knitr::kable(precis_managerR1, caption="Model estimates for the quadratic regression model with manager as unit of analysis, to test hypothesis H4 with the revised model excluding own team from the population predictors. Parameter a is the intercept, b1 is the linear coefficient and b2 the quadratic coefficient. Win rate is modelled as normally distributed with standard deviation sigma. See SI for priors.", digits = 2, booktabs = T, linesep="")
```
\newpage
# Revised models including additional opponent controls
The following results are for models that are identical to the preregistered original analyses presented in the main paper, but with two additional controls added to all models: (i) the relative strength of the opponent, calculated in the same way as the own team strength predictor, on the assumption that managers may alter their formations depending on opponent strength (e.g. play defensive formations against strong teams and attacking formations against weak teams); and (ii) the formation played by the opponent, coded as 4231 or non-4231, on the assumption that managers may attempt to counter 4231 with either the same formation, matching players in the same positions, or with a different formation in an attempt to break the 4231 domination. Including these controls had negligible effect on the parameter estimates, and did not qualitatively change conclusions regarding any of the hypotheses.
## Hypothesis H1: combination of population and personal information use
Table S17 shows almost identical model comparison for the revised models including additional opponent controls, compared to the original model given in the main text. Table S18 also shows very similar estimates for the full revised model compared to the original. The two additional control variables both reliably predict a manager's use of 4231: managers are more likely to use 4231 against stronger opponents, and when opponents use 4231 themselves.
```{r tableS17, echo = F, message = F}
mcomparisonR2 <- as.matrix(readRDS(file="modeloutputs/mcomparisonR2.rds"))
row.names(mcomparisonR2)[row.names(mcomparisonR2) == "m1R2"] <- "Null model"
row.names(mcomparisonR2)[row.names(mcomparisonR2) == "m2R2"] <- "Personal model"
row.names(mcomparisonR2)[row.names(mcomparisonR2) == "m3R2"] <- "Population model"
row.names(mcomparisonR2)[row.names(mcomparisonR2) == "m4R2"] <- "Full model"
knitr::kable(mcomparisonR2, caption="Model comparison to test hypothesis H1, with revised models including additional opponent controls.", digits = 2, booktabs = T, linesep="")
```
```{r tableS18, echo = F, message = F}
precis_m4R2 <- as.matrix(readRDS(file="modeloutputs/precis_m4R2.rds"))
# add rows, sub-headings for fixed and varying effects
precis_m4R2 <- rbind(fixed_effects = rep(NA,6), precis_m4R2)
precis_m4R2 <- rbind(precis_m4R2[1:14,], rep(NA,6), varying_effects = rep(NA,6), precis_m4R2[15:20,])
row.names(precis_m4R2) <- c("Fixed effects:", "Home/away: Home", "Home/away: Away", "Team strength", "Opponent strength", "Opponent 4231", "Personal 4231 use", "Personal 4231 win rate", "Personal 4231 use * personal 4231 win rate", "Personal 4231 use * personal win rate", "Population 4231 use", "Population 4231 win rate", "Population 4231 use * population 4231 win rate", "Population 4231 use * personal win rate", " ", "Varying effects:", "Manager", "Manager * personal 4231 use", "Manager * population 4231 win rate", "Division", "Division * personal 4231 use", "Division * population 4231 win rate")
options(knitr.kable.NA = '')
knitr::kable(precis_m4R2[,1:4], caption="Parameter estimates for the full revised model including additional opponent controls. Home/away is an indicator trait with separate estimates for formations used home and away. Varying effects show the standard deviations of the varying intercepts and slopes. See SI for full model specification and priors.", digits = 2, booktabs = T, linesep="")
```
## Hypothesis H2: ratio of population to personal 4231 use
For the revised model, the mean population:personal use ratio was -0.53, 89%CI[0.20, 1.32]. The mean is negative but this is unreliable and due to an extreme negative value. The confidence intervals, however, are similar to the original model, and widely encompass one indicating no support for the hypothesis.
## Hypothesis H3: variation across managers and divisions
Table S19 shows that for the revised model with additional opponent controls, there is more varation in 4231 personal and population use across managers than randomised data, and more variation across divisions in personal 4231 use, but not across divisions in population 4231 use. This is the same as the original model.
```{r tableS19, echo = F, message = F}
precis_sigmadiffsR2 <- as.matrix(readRDS(file="modeloutputs/precis_sigmadiffsR2.rds"))
row.names(precis_sigmadiffsR2)[row.names(precis_sigmadiffsR2) == "diff_pers_manager"] <- "Manager * personal 4231 use"
row.names(precis_sigmadiffsR2)[row.names(precis_sigmadiffsR2) == "diff_pop_manager"] <- "Manager * population 4231 use"
row.names(precis_sigmadiffsR2)[row.names(precis_sigmadiffsR2) == "diff_pers_division"] <- "Division * personal 4231 use"
row.names(precis_sigmadiffsR2)[row.names(precis_sigmadiffsR2) == "diff_pop_division"] <- "Division * population 4231 use"
knitr::kable(precis_sigmadiffsR2, caption="Tests of the differences between varying effects from the real data and varying effects from randomised data, to test hypothesis H3 with the revised model including additional opponent controls. Values shown are real minus randomised standard deviations.", digits = 2, booktabs = T, linesep="")
```
## Hypothesis H4: population to personal use ratio and win rate
Table S20 shows that there is no quadratic or linear relationship between manager win rate and population:personal use ratio for the revised model including additional opponent controls, just as for the original model.
```{r tableS20, echo = F, message = F}
precis_managerR2 <- as.matrix(readRDS(file="modeloutputs/precis_managerR2.rds"))
knitr::kable(precis_managerR2, caption="Model estimates for the quadratic regression model with manager as unit of analysis, to test hypothesis H4 with the revised model including additional opponent controls. Parameter a is the intercept, b1 is the linear coefficient and b2 the quadratic coefficient. Win rate is modelled as normally distributed with standard deviation sigma. See SI for priors.", digits = 2, booktabs = T, linesep="")
```
\newpage
# References