ggalluvial - sort the order of strata

Question

I've been trying for days now to sort the order of strata and flows in ggalluvial. I want to visualize the flow of patients through different screenings procedures (X1, X2, X3, X4) and color the flow based on the final diagnosis (Values in X4).

Can you help me sort the values within the groups in the first columns of example A and B? I want all red, yellow, and blue values stacked on top of each other within each of the groups.

So far I have tried various combinations of wide-format, aes.flow "backwards" and "forwards," lode.guidance, and lode.ordering...

If this is not possible in ggalluvial but possible in other packages, I'd like to know as well.

Thanks in advance.

DATA in wide format:

set.seed(1)
data <- tibble(
  ID = 1:879,
  X1 = sample(c("only_parent", "parent_and_3D", "only_3D"), size = 879, replace = TRUE, prob = c(0.1, 0.8, 0.1))) %>% 
  mutate(
    X2 = case_when(
      X1 == "only_parent" ~ sample(c("only_I", "not_identified"), size = n(), prob = c(0.1, 0.9), replace = TRUE),
      X1 == "parent_and_3D" ~ sample(c("only_I", "both_I_and_II", "only_II", "not_identified"), size = n(), prob = c(0.05, 0.05, 0.2, 0.7), replace = TRUE),
      X1 == "only_3D"~ sample(c("only_II", "not_identified"), size = n(), prob = c(0.1, 0.9), replace = TRUE),
      TRUE ~ NA_character_)) %>% 
  mutate(
    X3 = case_when(
      X2 == "only_I" ~ "PO_only",
      X2 == "both_I_and_II" ~ sample(c("PO_and_EHL", "PO_and_F/T", "PO_and_F/T_and_EHL"), size = n(), prob = c(0.3, 0.5, 0.2), replace = TRUE),
      X2 == "only_II"~ sample(c("F/T", "F/T_and_EHL", "EHL"), size = n(), prob = c(0.1, 0.6, 0.4), replace = TRUE),
      X2 == "not_identified" ~ "not_identified",
      TRUE ~ NA_character_)) %>% 
  mutate(
    X4 = case_when(
      X3 == "PO_only"    ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.02, 0.1, 0.88), replace = TRUE),
      X3 == "PO_and_EHL" ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.05, 0.2, 0.75), replace = TRUE),
      X3 == "PO_and_F/T" ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.05, 0.2, 0.75), replace = TRUE),
      X3 == "PO_and_F/T_and_EHL" ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.05, 0.2, 0.75), replace = TRUE),
      X3 == "F/T" ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.02, 0.1, 0.88), replace = TRUE),
      X3 == "F/T_and_EHL" ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.05, 0.2, 0.75), replace = TRUE),
      X3 == "EHL" ~ sample(c("Two_primary_ind", "One_primary_ind", "No TW"), size = n(), prob = c(0.02, 0.2, 0.88), replace = TRUE),
      X3 == "not_identified" ~ "not_identified",
      TRUE ~ NA_character_ ))

head(data)

# A tibble: 6 x 5
     ID X1            X2             X3             X4            
  <int> <chr>         <chr>          <chr>          <chr>         
1     1 parent_and_3D not_identified not_identified not_identified
2     2 parent_and_3D only_II        F/T_and_EHL    No TW         
3     3 parent_and_3D not_identified not_identified not_identified
4     4 only_parent   only_I         PO_only        No TW         
5     5 parent_and_3D only_II        F/T_and_EHL    No TW         
6     6 only_3D       not_identified not_identified not_identified

Example A
The values are not sorted in the bottom box of the first column.

data_long_a <- data %>% 
  group_by(X1, X2, X3, X4) %>% 
  count() %>% 
  mutate(
    fill_stat = factor(X4, levels = c("not_identified", "No TW", "One_primary_ind", "Two_primary_ind"))) %>% 
  ungroup  %>%
  arrange(fill_stat) %>% 
  mutate(subject = seq(1, n())) %>% 
  gather(key, value, -n , -subject, -fill_stat) %>% 
  mutate(
    key = factor(key, levels = c("X1", "X2", "X3", "X4"))) %>% 
  arrange(key, fill_stat) 



data_long_a %>% 
  filter(key %in% c("X1", "X2")) %>% 
  ggplot(
    aes(x = key,
        y = n,
        stratum = value, 
        alluvium = subject,
        label = value))+
  geom_flow(aes(fill = fill_stat)) +
  geom_stratum() +
  geom_text(stat = "stratum")+
  scale_fill_manual(values=c("#BAB3B3EB", "red", "yellow", "blue"))+
  theme_void()

Example B
The flow lines in the first column are not sorted.

data_long_b <- data %>%
  select(-X1) %>% 
  filter(X4 != "not_identified") %>% 
  group_by(X2, X3, X4) %>% 
  count() %>% 
  mutate(
    fill_stat = factor(X4, levels = c("not_identified", "No TW", "One_primary_ind", "Two_primary_ind"))) %>% 
  ungroup  %>%
  arrange(fill_stat) %>% 
  mutate(subject = seq(1, n())) %>% 
  gather(key, value, -n , -subject, -fill_stat) %>% 
  mutate(
    key = factor(key, levels = c("X2", "X3", "X4"))) %>% 
  arrange(key, fill_stat) 


data_long_b %>% 
  ggplot(
    aes(x = key,
        y = n,
        stratum = value, 
        alluvium = subject,
        label = value))+
  geom_flow(aes(fill = fill_stat),
            aes.flow = "backward") +
  geom_stratum() +
  geom_text(stat = "stratum")+
  scale_fill_manual(values=c("red", "yellow", "blue"))+
  theme_void()

Answer 1

The background here is that, even though the strata (the different values stacked at each axis) may have a natural order, the alluvia, which represent individual cases or cohorts, usually do not. This means that one job of a stat layer (e.g. stat_alluvium()) is to determine an ordering of the lodes within each stratum. (This then determines the flows between strata.)

To improve clarity, stat_alluvium() and stat_flow() use the strata of the cases or cohorts at nearby axes to guide their positioning at a given axis. By default, it does this in a "zigzag" order, adapted from the alluvial package; see the "lode guidance" documentation for additional options.

This behavior can be problematic when the user wants to group cohorts together within strata, as when lodes and flows are assigned aesthetics (usually fill, but optionally alpha, colour, linetype, and size). The aes.bind parameter addresses this problem by prioritizing aesthetics before (but not instead of) strata in nearby axes when determining lode orderings.

@Steen provided a syntactic answer, which i'll basically copy here. I make one change, from stat_flow() to stat_alluvium() in Example B, to illustrate that aes.bind can be passed to and will be correctly interpreted by either geom layer.

Example A:

data_long_a %>% 
  filter(key %in% c("X1", "X2")) %>% 
  ggplot(
    aes(x = key,
        y = n,
        stratum = value, 
        alluvium = subject,
        label = value))+
  geom_flow(aes(fill = fill_stat), aes.bind = TRUE) +
  geom_stratum() +
  geom_text(stat = "stratum")+
  scale_fill_manual(values=c("#BAB3B3EB", "red", "yellow", "blue"))+
  theme_void()

Example B:

data_long_b %>% 
  ggplot(
    aes(x = key,
        y = n,
        stratum = value, 
        alluvium = subject,
        label = value))+
  geom_alluvium(aes(fill = fill_stat),
                aes.bind = TRUE) +
  geom_stratum() +
  geom_text(stat = "stratum")+
  scale_fill_manual(values=c("red", "yellow", "blue"))+
  theme_void()

^{Created on 2019-07-27 by the reprex package (v0.2.1)}

Answer 2

Like Cory Brunson writes in the comment: "The key is aes.bind = TRUE "

Example A:

data_long_a %>% 
  filter(key %in% c("X1", "X2")) %>% 
  ggplot(
    aes(x = key,
        y = n,
        stratum = value, 
        alluvium = subject,
        label = value))+
  geom_flow(aes(fill = fill_stat), aes.bind = TRUE) +
  geom_stratum() +
  geom_text(stat = "stratum")+
  scale_fill_manual(values=c("#BAB3B3EB", "red", "yellow", "blue"))+
  theme_void()

Example B:

data_long_b %>% 
  ggplot(
    aes(x = key,
        y = n,
        stratum = value, 
        alluvium = subject,
        label = value))+
  geom_flow(aes(fill = fill_stat),
            aes.bind = TRUE) +
  geom_stratum() +
  geom_text(stat = "stratum")+
  scale_fill_manual(values=c("red", "yellow", "blue"))+
  theme_void()