Evaluation of team rank prediction in sports league by sum of residuals

J1_2019_pred_results<-read_csv(file.path("2019-12-29-evaluation-of-team-rank-prediction-in-sports-league-by-sum-of-residuals_files","日刊スポーツ・サッカー担当の19年J1順位予想と結果.csv"))

## Parsed with column specification:
## cols(
##   rank = col_double(),
##   Hamamoto = col_character(),
##   No = col_character(),
##   Hosaka = col_character(),
##   Shimoda = col_character(),
##   Sugiyama = col_character(),
##   Kinoshita = col_character(),
##   Okazaki = col_character(),
##   Matsuo = col_character(),
##   Iwata = col_character(),
##   Kamiya = col_character(),
##   Maeda = col_character(),
##   Sanefuji = col_character(),
##   Kikukawa = col_character(),
##   Ishikawa = col_character(),
##   Uehara = col_character(),
##   Ogishima = col_character(),
##   Results = col_character()
## )

#J1_2019_pred_results %>% DT::datatable()
J1_2019_pred_results %>% knitr::kable()

rank	Hamamoto	No	Hosaka	Shimoda	Sugiyama	Kinoshita	Okazaki	Matsuo	Iwata	Kamiya	Maeda	Sanefuji	Kikukawa	Ishikawa	Uehara	Ogishima	Results
1	Kashima	Kashima	Kawasaki_F	Kawasaki_F	Kashima	Urawa	Urawa	Kawasaki_F	Urawa	Kashima	Kawasaki_F	Urawa	Kawasaki_F	Kawasaki_F	Kawasaki_F	Kawasaki_F	Yokohama
2	Kawasaki_F	Tokyo	Kashima	Kashima	Kawasaki_F	Kawasaki_F	Kawasaki_F	Kashima	Kawasaki_F	Urawa	Kashima	Kobe	Kobe	Kobe	Kobe	Kashima	Tokyo
3	Kobe	Kawasaki_F	Sapporo	C_Osaka	Urawa	Tokyo	Tokyo	Nagoya	Kashima	G_Osaka	G_Osaka	Kashima	Tokyo	G_Osaka	Nagoya	Kobe	Kashima
4	Urawa	Kobe	Urawa	Sapporo	G_Osaka	C_Osaka	G_Osaka	Urawa	Kobe	Kawasaki_F	Kobe	Kawasaki_F	Kashima	Tokyo	Sapporo	Urawa	Kawasaki_F
5	G_Osaka	Urawa	Kobe	Sendai	Tokyo	Kobe	Sapporo	Sapporo	G_Osaka	Shimizu	Urawa	C_Osaka	G_Osaka	Urawa	Kashima	G_Osaka	C_Osaka
6	Tokyo	Shonan	G_Osaka	Urawa	Tosu	Kashima	Yokohama	G_Osaka	Nagoya	C_Osaka	Tokyo	G_Osaka	Sapporo	Kashima	G_Osaka	Nagoya	Hiroshima
7	Sapporo	Nagoya	Yokohama	Kobe	Sapporo	Nagoya	Kashima	Kobe	Sapporo	Tokyo	Nagoya	Tokyo	Urawa	Nagoya	Urawa	Hiroshima	G_Osaka
8	C_Osaka	Hiroshima	Tokyo	Iwata	Kobe	G_Osaka	Kobe	Shonan	C_Osaka	Nagoya	Iwata	Yokohama	Nagoya	Hiroshima	Tokyo	Tokyo	Kobe
9	Shonan	G_Osaka	C_Osaka	Tokyo	Sendai	Sapporo	Hiroshima	Yokohama	Tokyo	Hiroshima	Sapporo	Sapporo	Shimizu	Sapporo	Tosu	Iwata	Ooita
10	Shimizu	Yokohama	Shonan	Ooita	Hiroshima	Hiroshima	Nagoya	Tosu	Iwata	Iwata	Yokohama	Nagoya	Hiroshima	Sendai	Hiroshima	Sapporo	Sapporo
11	Sendai	C_Osaka	Iwata	Shonan	Shimizu	Tosu	Iwata	Sendai	Tosu	Kobe	Shimizu	Hiroshima	Sendai	C_Osaka	C_Osaka	Shimizu	Sendai
12	Hiroshima	Sapporo	Shimizu	Yokohama	Yokohama	Sendai	Shimizu	Hiroshima	Shimizu	Sendai	C_Osaka	Shimizu	Shonan	Yokohama	Yokohama	Yokohama	Shimizu
13	Yokohama	Iwata	Hiroshima	Shimizu	Shonan	Shonan	Shonan	Tokyo	Shonan	Yokohama	Hiroshima	Sendai	Yokohama	Shimizu	Shonan	C_Osaka	Nagoya
14	Iwata	Tosu	Nagoya	Hiroshima	C_Osaka	Shimizu	C_Osaka	C_Osaka	Yokohama	Tosu	Tosu	Tosu	C_Osaka	Shonan	Sendai	Shonan	Urawa
15	Nagoya	Matsumoto	Tosu	G_Osaka	Matsumoto	Iwata	Sendai	Ooita	Hiroshima	Matsumoto	Sendai	Iwata	Iwata	Tosu	Iwata	Sendai	Tosu
16	Tosu	Sendai	Sendai	Nagoya	Iwata	Yokohama	Matsumoto	Shimizu	Sendai	Sapporo	Shonan	Shonan	Tosu	Matsumoto	Shimizu	Tosu	Shonan
17	Matsumoto	Shimizu	Ooita	Tosu	Nagoya	Matsumoto	Tosu	Iwata	Matsumoto	Shonan	Matsumoto	Matsumoto	Ooita	Iwata	Matsumoto	Matsumoto	Matsumoto
18	Ooita	Ooita	Matsumoto	Matsumoto	Ooita	Ooita	Ooita	Matsumoto	Ooita	Ooita	Ooita	Ooita	Matsumoto	Ooita	Ooita	Ooita	Iwata

Japanese professional soccor league is called “J league”" and its top level is J1. Juset before season starts sports media predict team rank. It is unusual the media review their prediction after finishing the season, but Nikkansports.com did this year (Table above). In that article their evaluation of the prediction done by 17 writers is rough and not sufficiently analytical, so that I tried to analyze the data here for fun. No writers predicted that Yokohama won this season. I am just curious who’s prediction is very close to the result and which temas did surprised performance.

convert data format

Results<-J1_2019_pred_results %>% select(rank,Results) %>% rename(team=Results,results=rank)
J1_2019_pred_results.mod<-J1_2019_pred_results %>% select(-Results) %>% gather("person","team",-1) %>% rename(prediction=rank) %>% left_join(Results,by="team")

Who is the best predictor?

J1_2019_pred_results.mod %>% mutate(var=(prediction-results)^2) %>% group_by(person) %>% summarise(`sum of residuals`=sum(var)) %>% arrange(`sum of residuals`)

## # A tibble: 16 x 2
##    person    `sum of residuals`
##    <chr>                  <dbl>
##  1 Sanefuji                 410
##  2 Ishikawa                 442
##  3 Hosaka                   482
##  4 Kikukawa                 482
##  5 Hamamoto                 506
##  6 Okazaki                  508
##  7 No                       528
##  8 Maeda                    554
##  9 Sugiyama                 574
## 10 Shimoda                  590
## 11 Kinoshita                592
## 12 Ogishima                 594
## 13 Uehara                   604
## 14 Kamiya                   614
## 15 Matsuo                   682
## 16 Iwata                    754

Smaller sum of resuduals means there are less differences beteen their prediction and result. Congratulations Mr. Sanefuji (correct pronounciation?).

Which teams are surprise?

J1_2019_pred_results.mod %>% mutate(var=(prediction-results)^2) %>% group_by(team) %>% summarise(`sum of residuals`=sum(var)) %>% arrange(desc(`sum of residuals`))

## # A tibble: 18 x 2
##    team       `sum of residuals`
##    <chr>                   <dbl>
##  1 Yokohama                 1767
##  2 Urawa                    1746
##  3 Ooita                    1137
##  4 Iwata                     590
##  5 C_Osaka                   585
##  6 Nagoya                    540
##  7 Hiroshima                 431
##  8 Tokyo                     425
##  9 Shonan                    344
## 10 Kobe                      271
## 11 Sapporo                   258
## 12 Tosu                      189
## 13 Sendai                    179
## 14 G_Osaka                   155
## 15 Shimizu                   128
## 16 Kawasaki_F                 93
## 17 Kashima                    60
## 18 Matsumoto                  18

Across teams I calculated the sum of residuals. In this analysis larger value means more surprise for those writers. Yokohama F. Marinos gained the largest value, which means writers underestimate Yokohama’s performance. On the other hand the second largest value of Urwawa has different meaning. Although Urwa Red Diamonds became a finalist in 2019 AFC (Asian Football Confederation) Champions League, their performance was so bad that the team head coach had to quit in the middle of the season. Kashima Antlers was the winner of 2018 AFC Champions League Final, so that many writers predicted its very good performance. As expected Kashima did good job (3rd place) and its sum of resisuals are very small. Unfortunately Matsumoto Yamaga’s performance was as expected in a bad meaning The team is too small to be competitive in the top level league. I hope this team in a small town has learned a lot from this year’s experiences.

Conclusion

There could be well established method for this kind of analysis, but I am satisfied with my rough analysis using sum of residuals.