Home / Questions / 554918
Accepted
kev
Asked: 2022-09-02 13:25:25 +0800 CST 2022-09-02 13:25:25 +0800 CST

How to perform regression estimation in a for in R?

  • 772

I have the following database: enter image description here

I need to estimate regressions of the following form:

ir_total = b0 + b1 9dejulio + b2 Azul + b3 Bolivar
ir_total = b0 + b1 9dejulio + b2 Azul + b3 CoronelSuarez
.
.
.
.
ir_total = b0 + b1 pehuajo + b2 Dolores + b3 Junin

That is, I need to estimate regressions with the variable y=ir_total, 3 regressors that will go through all the different departments. Finally print in a DataFrame or table the coefficients and the R^2 of each regression

I am trying something like the following

n=ncol(data)

for (i in 3:n) {
    model = lm (data = data, formula = data[,2] ~ data[,i] + data [,i+1] + data[,i+2])

}
r

2 Answers

  1. Best Answer

    First, I synthetically generate a data set similar to the one you show:

    set.seed(2022)
df <- data.frame(year=sample(2001:2018, 10, replace=TRUE),
                 ir_total=runif(10),
                 Azul=runif(10),
                 Bolivar=runif(10),
                 Dolores=runif(10),
                 Junin=runif(10)
                 )

    The first problem is to generate all the combinations of 3 independent variables, for this we select the column names that are going to participate in the calculation, in your case, all except the first two and then we simply combn()generate the combinations:

    independientes <- colnames(df[, c(-1,-2)])
combinaciones <- apply(combn(independientes, m=3), 2, 
                       FUN=function(x){paste0("`", x, "`", collapse = ' + ')})
combinaciones
    
[1] "Azul + Bolivar + Dolores"  "Azul + Bolivar + Junin"   
[3] "Azul + Dolores + Junin"    "Bolivar + Dolores + Junin"

    We end up with partial strings, but we still need to complete the formula and actually turn them into a formula:

    combinaciones <- sapply(paste("ir_total ~", combinaciones), as.formula)

    We now have a list of formulas, we simply lapply()apply lmthem to each element:

    lapply(combinaciones, lm, df)

    Finally, having a list with all the models, it is relatively easy to "extract" the data you mention

    lapply(modelos, FUN=function(x) {cbind(r2=summary(x)$r.squared,
                                       coef1=coef(x)[2],
                                       coef2=coef(x)[3],
                                       coef3=coef(x)[4]
                                       )}) -> datos

datos <- do.call(rbind, datos)
rownames(datos) <- NULL

datos

             r2      coef1       coef2      coef3
[1,] 0.18734057  0.2333279 -0.50774617  0.0534951
[2,] 0.23616736  0.1888734 -0.74079657 -0.3212276
[3,] 0.02565005  0.1438587  0.07310115  0.1276001
[4,] 0.20258716 -0.7262850 -0.08811687 -0.4056673
    • 1

  2. Perhaps this is an option to address this question. It may not be the most efficient, but it could help you. Considering that there is no sample of the data, here is an example with fictitious data:

    set.seed(123)                         
y <- rnorm(1000)
x1 <- rnorm(1000) + 0.123 * y
x2 <- rnorm(1000) + 0.123 * x1 + 0.456 * y
x3 <- rnorm(1000) - 0.123 * x1 + 0.456 * x2 - 0.789 * y
data <- data.frame(y, x1, x2, x3)

# Crear una lista vacía  
modelos_resultados <- list()                 

# Crear for-loop
for(i in 2:ncol(data)) {                 
  
  predictores_i <- colnames(data)[2:i]    # Crear vector con las variables x
  modelos_resultados [[i - 1]] <- summary(     # Guardar resultados de cada modelo en la lista
    lm(y ~ ., data[ , c("y", predictores_i )]))
  
}

# Extraer los resultados de los modelos en formato tidy
resultados_tidy <- purrr::map(modelos_resultados, broom::tidy) |> 
  dplyr::bind_rows(.id="predvar") |> 
  filter(predvar  %in% c(1:3)) # Filtrar por el número de modelos que necesitas, en este caso, considera los primeros tres modelos. 

# Extraer r^2 de cada modelo
r.squared <- map_dfr(modelos_resultados, ~ glance(.x) |> 
                       select(r.squared)) |> mutate(Modelo_ID = as.character(1:3))

# Agregar r^2 al resto de las variables de cada modelo.  
resultados_tidy <- resultados_tidy |> 
left_join(r.squared, by = c("predvar" = "Modelo_ID")) |> 
  select(predvar, everything())

    DATAFRAME

                   y           x1            x2          x3
1   -0.560475647 -1.064737229 -0.8981432962  0.01331714
2   -0.230177489 -1.068266875  0.0005801183 -0.01448573
3    1.558708314  0.173740882  0.1905519481 -2.61246459
4    0.070508391 -0.123502601  1.2361886536 -0.17402286
5    0.129287735 -2.533440383 -0.0785220848  2.77228931
6    1.715064987  1.251526449  0.3207390712 -1.39828203
7    0.460916206  0.306418429 -1.5590257073 -0.19877619
8   -1.265061235  2.260604841 -0.9424946334  0.10577487
9   -0.686852852  0.600715337  1.8067019382  1.90171929

    OUTPUT

      predvar term        estimate std.error statistic  p.value r.squared
  <chr>   <chr>          <dbl>     <dbl>     <dbl>    <dbl>     <dbl>
1 1       (Intercept)  0.00739    0.0307     0.240 8.10e- 1    0.0415
2 1       x1           0.197      0.0299     6.57  7.93e-11    0.0415
3 2       (Intercept)  0.0138     0.0283     0.487 6.27e- 1    0.190 
4 2       x1           0.112      0.0282     3.97  7.60e- 5    0.190 
5 2       x2           0.359      0.0266    13.5   2.37e-38    0.190 
6 3       (Intercept)  0.00563    0.0228     0.247 8.05e- 1    0.473 
7 3       x1           0.0140     0.0232     0.604 5.46e- 1    0.473 
8 3       x2           0.452      0.0218    20.7   1.96e-79    0.473 
9 3       x3          -0.429      0.0186   -23.1   6.80e-95    0.473
    • 0