Strings As Factors

Disclaimer

This notebooks tries to show the difference between the two datatypes ‘strings’ and ‘factors’ in R.

Data

The following dataset is used. This data set is based on the data used in exercise 5.

### # read tibble from file
s_wwg_path <- 'https://charlotte-ngs.github.io/lbgfs2020/misc/weaningweightbeef.csv'
tbl_beef_data <- readr::read_csv(file = s_wwg_path)

Parsed with column specification:
cols(
  Animal = col_double(),
  Sire = col_double(),
  Dam = col_double(),
  Herd = col_double(),
  `Weaning Weight` = col_double()
)

tbl_beef_data

l_sex <- list(abbrev = c('F', 'M'), name = c('female', 'male'))

We add a column ‘sex’ to the data where F stands for a female animal and M stands for a male animal, respectively.

tbl_beef_data$Sex <- sample(l_sex$abbrev, size = nrow(tbl_beef_data), replace = TRUE)
tbl_beef_data

Based on the above output, we see that the column ‘Sex’ is a character, hence a ‘string’. If this data is read again

s_ex_beef_data <- 'extended_beef_data.csv'
readr::write_csv2(tbl_beef_data, path = s_ex_beef_data)
df_ex_beef_str <- read.csv2(file = s_ex_beef_data)
df_ex_beef_str

Changing the option of ‘stringsAsFactors’

df_ex_beef_fct <- read.csv2(file = s_ex_beef_data, stringsAsFactors = TRUE)
df_ex_beef_fct

Linear Model

Using the above data in a linear model.

lm_str <- lm(Weaning.Weight ~ Sex + Herd, data = df_ex_beef_str)
summary(lm_str)

Call:
lm(formula = Weaning.Weight ~ Sex + Herd, data = df_ex_beef_str)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.46301 -0.14199 -0.01330  0.09358  0.64699 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)  2.63972    0.35089   7.523 4.36e-06 ***
SexM        -0.12864    0.20312  -0.633    0.538    
Herd        -0.05807    0.19667  -0.295    0.772    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.3368 on 13 degrees of freedom
Multiple R-squared:  0.03003,   Adjusted R-squared:  -0.1192 
F-statistic: 0.2012 on 2 and 13 DF,  p-value: 0.8202

lm_fct <- lm(Weaning.Weight ~ Sex + Herd, data = df_ex_beef_fct)
summary(lm_fct)

Call:
lm(formula = Weaning.Weight ~ Sex + Herd, data = df_ex_beef_fct)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.46301 -0.14199 -0.01330  0.09358  0.64699 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)  2.63972    0.35089   7.523 4.36e-06 ***
SexM        -0.12864    0.20312  -0.633    0.538    
Herd        -0.05807    0.19667  -0.295    0.772    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.3368 on 13 degrees of freedom
Multiple R-squared:  0.03003,   Adjusted R-squared:  -0.1192 
F-statistic: 0.2012 on 2 and 13 DF,  p-value: 0.8202

Herd

Add an additional Herd

The same model

lm_fct_hd <- lm(Weaning.Weight ~ Sex + Herd, data = df_ex_beef_fct_hd)
summary(lm_fct_hd)

Call:
lm(formula = Weaning.Weight ~ Sex + Herd, data = df_ex_beef_fct_hd)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.45414 -0.13178 -0.01757  0.07147  0.65586 

Coefficients:
             Estimate Std. Error t value Pr(>|t|)    
(Intercept)  2.550655   0.325482   7.837  2.8e-06 ***
SexM        -0.101690   0.216483  -0.470    0.646    
Herd        -0.004828   0.153718  -0.031    0.975    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.3379 on 13 degrees of freedom
Multiple R-squared:  0.0236,    Adjusted R-squared:  -0.1266 
F-statistic: 0.1571 on 2 and 13 DF,  p-value: 0.8562

Changing the datatype of the Herd into a factor

df_ex_beef_fct_hd$Herd <- as.factor(df_ex_beef_fct_hd$Herd)
df_ex_beef_fct_hd

NA

lm_fct_hd <- lm(Weaning.Weight ~ Sex + Herd, data = df_ex_beef_fct_hd)
summary(lm_fct_hd)

Result

For all columns which are of a numeric type (integer or double), the lm() function in R fits a regression. For all columns which are of type factor, the lm() function uses the variable as a fixed effect.