Os dados descrevem 5000 encontros relâmpagos (speed dating, procura no google) de 4 minutos envolvendo 310 jovens americanos. Os dados originais foram coletados por professores da Columbia Business School no experimento descrito aqui. Nesta análise estamos usando uma versão com menos atributos que pode ser encontrado aqui. Os atributos selecionado são:

Os participantes tinham vários encontros de 4 minutos por noite. Após cada um, preenchiam fichas avaliando aqueles com quem se encontraram. Cada linha nos dados representa um desses encontros.

Nesta análise vamos utilizar regressão logística em um conjunto nestas 4 variáveis explicativas para responder quais fatores nos dados têm efeito relevante na chance do casal ter um match?

library(tidyverse)
library(tidyr)
library(stargazer)
library(caret)
library(GGally)
library(broom)

speed_dating = read_csv("dados/speed-dating2.csv") %>% 
  select(dec, gender, age, age_o, fun, attr, fun) %>% 
  mutate(diff_age = age - age_o,
         dec = case_when( .$dec == "no" ~ 0,
                          .$dec == "yes" ~ 1)) %>% 
  select(-age, -age_o)

Vamos explorar os dados…

ggpairs(speed_dating)

Percebemos que existe correlação entre algumas variáveis. Vamos criar o modelo!

logit <- glm(dec ~ gender + fun + attr + diff_age, data = speed_dating, family = "binomial")
tidy(logit, conf.int = TRUE, conf.level = 0.95)
glance(logit)

Utilizamos Regressão Logisticaa para analisar se as variáveis gender, fun, attr e diff_age tem uma associação na chance do casal ter um match. Os resultados da regressão indicam que um modelo no formato chance do casal ter um match = -6 + 0.22 * gender + 0.29 * fun + 0.55 * attr - 0 * diff_age apresenta AIC: 4881.2. A chance do casal ter um match tem relação com attr de (b = [0.50361597; 0.60356618], IC com 95%). O aumento de 1 unidade na variável attr produz uma mudança de aproximadamente 1/2 unidade na chance do casal ter um match. As variáveis gender e fun possuem relação semelhante a chance do casal ter um match de (b = [0.08835595;0.36457205] e b = [0.24822203 0.33859352], IC com 95%), respectivamente. O aumento de 1 unidade nessas variáveis produz uma mudança de aproximadamente 1/4 unidade na chance do casal ter um match. A diferenca de idade entre os participantes, representada por attr, possui relação despresível.

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