Introduction

In the PUMA survey income was included as monthly personal gross income across 15 categories ranging from up to 250€ to more than 6000€.

Frequency personal income

table(WaveOne$SD15_Perseink)

Recoding

For the analysis, the scale was recoded into four categories, namely up to 1300€, 1301€ to 2500€, 2501€ to 4000€ and more than 4000€ to provide meaningful categories on sample composition.

In the first step a new variable with missings is defined:

WaveOne$inc <- NA

Recode the lowest five categories capturing all income up to 1.300€ into the new category up to 1.300€:

WaveOne$inc[WaveOne$SD15_Perseink %in% 
              levels(WaveOne$SD15_Perseink)[3:7]] <- "up to 1.300€" 

Recode the next four categories capturing all income from 1.301 to 2.500€ into the new category 1.301 to 2.500€:

WaveOne$inc[WaveOne$SD15_Perseink %in% 
              levels(WaveOne$SD15_Perseink)[8:11]] <- "1.301 to 2.500€"

Contine with the next three categories covering income from 2.501 to 4.000€ into the category 2.501 to 4.000€:

WaveOne$inc[WaveOne$SD15_Perseink %in% 
              levels(WaveOne$SD15_Perseink)[12:14]] <- "2.501 to 4.000€"

And summarize the highest income categories from 4.001 to more than 6.000€ into the category more than 4.000€:

WaveOne$inc[WaveOne$SD15_Perseink %in% 
              levels(WaveOne$SD15_Perseink)[15:17]] <- "more than 4.000€" 

Finally, confirm all Don't know and No answer as missings:

WaveOne$inc[WaveOne$SD15_Perseink %in% 
              levels(WaveOne$SD15_Perseink)[1:2]] <- NA # Missings

And format the new variable as factor with specific levels:

WaveOne$inc <- factor(WaveOne$inc, 
                      levels = c("up to 1.300€", 
                                 "1.301 to 2.500€", 
                                 "2.501 to 4.000€", 
                                 "more than 4.000€"))

Frequency categorised income

Now take a look at the new variable:

table(WaveOne$inc)

Comparison original vs. paper

The following charts compare the original coding against the coding applied in the article (Table 2, p.11).

p1 <- WaveOne %>%
  # filter all contacted by telephone
  filter(is.na(PUMA1)==FALSE) %>% 
  # filter all receiving incentive
  filter(PUMA1==levels(WaveOne$PUMA1)[1] | PUMA1==levels(WaveOne$PUMA1)[2]) %>% 
  # filter nonresponse
  filter(valid==1) %>%
  ggplot(aes(SD15_Perseink)) + 
  geom_bar(width=0.8) +
  geom_vline(xintercept = 2.5) +
  labs(x="income categories",
       title = "Original Coding") +
  theme(axis.text.x = element_text(angle = 60, hjust = 1),
        plot.title = element_text(hjust = 0.5))

p2 <- WaveOne %>%
  # filter all contacted by telephone
  filter(is.na(PUMA1)==FALSE) %>% 
  # filter all receiving incentive
  filter(PUMA1==levels(WaveOne$PUMA1)[1] | PUMA1==levels(WaveOne$PUMA1)[2]) %>% 
  # filter nonresponse
  filter(valid==1) %>%
  # drop missings for new income variable
  drop_na(inc) %>%
  ggplot(aes(inc)) + 
  geom_bar() +
  labs(x="income categories",
       title = "Paper Coding") +
  theme(axis.text.x = element_text(angle = 60, hjust = 1),
        plot.title = element_text(hjust = 0.5))

grid.arrange(p1,p2, ncol=2)

Income quintils

For the nonresponse analysis, the income quintiles from the micro-census data are used.