Skip to contents

A snake plot makes three kinds of design decision on the reader’s behalf. The colour scale decides whether a response scale is read as two opposing sides around a neutral midpoint or as a single ordered intensity. The reading order decides in which direction each band is read and on which side the first band starts. The geometry (band height, gap and width) decides the aspect ratio of the figure and how much of the page it takes. Each decision has a default, and each default is wrong for some data.

This guide covers the built-in palettes (snake_palette(), snake_palettes), how many colours a plot uses, the difference between survey_sequence() and sequential_dist(), the ordering of items (sort_by), the reading order (flow, start_from), the label position (label_align), and the geometry (band_height, band_gap, plot_width). Reading order for time-ordered data is also shown in the sequences guide and the activity guide.

Data

ema_emotions holds the person-level emotion ratings of the students in an experience sampling study (Neubauer & Schmiedek, 2024; data CC-BY 4.0 at https://osf.io/bhq3p): each student’s mean rating of each emotion over the study, rounded to the 1–5 scale.

  • Respondent: one row per student.
  • Items: Happy, Afraid, Sad, Balanced, Exhausted, Cheerful, Worried, Lively, Angry, Relaxed — how strongly the emotion was felt, 1 (not at all) to 5 (extremely).
dim(ema_emotions)
#> [1] 280  10
head(ema_emotions)
#>   Happy Afraid Sad Balanced Exhausted Cheerful Worried Lively Angry Relaxed
#> 1     4      2   2        4         3        3       3      3     1       4
#> 2     4      1   2        3         4        4       2      3     1       3
#> 3     3      1   2        4         4        3       2      2     1       4
#> 4     4      1   1        3         3        4       2      2     1       3
#> 5     4      2   2        4         2        4       2      3     2       3
#> 6     2      2   2        3         2        2       2      2     2       3

The data hold 280 respondents and 10 items. The legends below show the response levels as the numbers 1 to 5.

The built-in palettes

snake_palettes is a named list of ten palettes of seven anchor colours each; snake_palette() returns one of them interpolated to n colours.

names(snake_palettes)
#>  [1] "classic" "earth"   "ocean"   "sunset"  "berry"   "blues"   "greens"  "grays"  
#>  [9] "warm"    "viridis"
snake_palette("earth", n = 5)
#> [1] "#8C510A" "#CFA155" "#C7C7C7" "#5AB2A7" "#01665E"
snake_palette("blues", n = 5)
#> [1] "#EFF3FF" "#B2D2E8" "#6BAED6" "#3181BD" "#084594"

The five-colour earth palette runs from brown through a grey midpoint (#C7C7C7) to teal: a diverging palette. The five-colour blues palette runs from near-white to dark blue: a sequential palette. The difference can be measured rather than judged by eye. The CIE L* lightness of a colour runs from 0 (black) to 100 (white); the next chunk converts every palette, interpolated to five colours, to L* with grDevices::convertColor() and tabulates it by palette and level.

swatches <- data.frame(
  palette = rep(names(snake_palettes), each = 5),
  level   = rep(seq_len(5), times = length(snake_palettes)),
  colour  = unlist(lapply(names(snake_palettes), snake_palette, n = 5))
)
lab <- with(swatches, convertColor(t(col2rgb(colour)) / 255,
                                   from = "sRGB", to = "Lab"))
swatches <- cbind(swatches, round(lab))
xtabs(L ~ palette + level, data = swatches)
#>          level
#> palette    1  2  3  4  5
#>   berry   45 71 85 77 54
#>   blues   96 83 68 52 30
#>   classic 39 65 84 67 42
#>   earth   40 69 80 67 38
#>   grays   97 82 62 42 15
#>   greens  96 85 73 57 33
#>   ocean   48 69 89 76 49
#>   sunset  48 71 86 61 27
#>   viridis 91 77 64 49 36
#>   warm    97 82 70 56 33

Two families are visible in the table. In berry, classic, earth, ocean and sunset the lightness rises from level 1 to the grey level 3 (L* 80 to 89) and falls again towards level 5: the two ends are dark and the middle is light, so distance from the midpoint is carried by lightness and the side of the midpoint by hue. In blues, grays, greens, viridis and warm the lightness falls monotonically from level 1 (L* 91 to 97) to level 5 (L* 15 to 36), so the order of the levels is carried by lightness alone. The arms of the diverging palettes are not all balanced: in sunset level 1 has L* 48 and level 5 has L* 27, so the high end is darker than the low end, whereas earth (40 against 38), classic (39 against 42) and ocean (48 against 49) are close to symmetric. This guide makes no claim about how any palette appears to readers with a colour-vision deficiency; that would require a simulation that is not shown here.

Pass the number of levels

survey_snake(), survey_sequence() and sequential_dist() use only the first length(levels) colours of colors. snake_palette() returns seven colours by default, so a five-level scale given the default palette uses the first five of them.

snake_palette("earth")
#> [1] "#8C510A" "#BF812D" "#DFC27D" "#C7C7C7" "#80CDC1" "#35978F" "#01665E"
head(snake_palette("earth"), 5)
#> [1] "#8C510A" "#BF812D" "#DFC27D" "#C7C7C7" "#80CDC1"
snake_palette("earth", n = 5)
#> [1] "#8C510A" "#CFA155" "#C7C7C7" "#5AB2A7" "#01665E"

The first five of the seven earth colours end with the grey midpoint #C7C7C7 at position 4 and the light teal #80CDC1 at position 5: on a five-level scale the neutral colour would be assigned to level 4 (quite) and the darkest teal would never be used. With n = 5 the grey is at position 3, the midpoint of the scale, and both ends are the darkest colours.

survey_sequence(ema_emotions,
                colors = snake_palette("earth"),
                title = "Seven colours on a five-level scale")

survey_sequence(ema_emotions,
                colors = snake_palette("earth", n = 5),
                title = "snake_palette(\"earth\", n = 5)")

The two plots draw the same counts; only the colour assigned to each level differs. In the first, level 3 is drawn in the light sand colour and level 4 in grey; in the second, level 3 is grey and the scale is symmetric around it.

Scales with and without a neutral midpoint

A diverging palette states that the middle level is neutral and that the two ends are opposites. That is true of an agreement scale (disagree, neutral, agree) but not of an intensity scale: rating anger as Moderately is not a neutral point between two kinds of anger, but more anger than Slightly. The emotion items are intensity ratings. The share of respondents at each level of each item:

round(100 * prop.table(xtabs(~ ind + values, data = stack(ema_emotions)),
                       margin = 1))
#>            values
#> ind          1  2  3  4  5
#>   Happy      1  9 25 64  1
#>   Afraid    49 41  6  4  0
#>   Sad       43 48  6  3  0
#>   Balanced   1 11 31 54  3
#>   Exhausted  3 45 36 15  1
#>   Cheerful   2 13 31 53  1
#>   Worried   20 55 13 11  1
#>   Lively     4 32 45 19  0
#>   Angry     70 25  4  1  0
#>   Relaxed    1 18 32 49  1

The negative emotions are concentrated at the bottom of the scale: 70% of the respondents rated Angry at level 1 and 1% above level 3, and Afraid and Sad have 49% and 43% at level 1. The positive emotions are concentrated at level 4 (64% for Happy, 53% for Cheerful, 54% for Balanced). For such a scale, sequential_dist() draws the same stacked bars as survey_sequence() with a single-hue ramp, so a darker segment means a stronger emotion.

sequential_dist(ema_emotions)

The default ramp is a blue (hue = 210); hue selects another hue, and colors accepts any five colours, for example snake_palette("greens", n = 5). survey_sequence() with a diverging palette is the right choice for agreement scales and other bipolar scales; sequential_dist() for intensity and frequency scales.

Ordering the items

The input order of the items is rarely the order in which a reader wants to compare them. sort_by = "mean" orders the bands by ascending item mean; sort_by = "net" by the share of responses above the scale midpoint minus the share below it. Both scores, computed directly and sorted with base R’s sort_by():

item_scores <- data.frame(
  item = names(ema_emotions),
  mean = colMeans(ema_emotions),
  net  = colMeans(ema_emotions > 3) - colMeans(ema_emotions < 3)
)
sort_by(item_scores, ~ mean)
#>                item mean    net
#> Angry         Angry 1.37 -0.936
#> Afraid       Afraid 1.65 -0.864
#> Sad             Sad 1.70 -0.871
#> Worried     Worried 2.19 -0.629
#> Exhausted Exhausted 2.66 -0.314
#> Lively       Lively 2.79 -0.168
#> Relaxed     Relaxed 3.30  0.304
#> Cheerful   Cheerful 3.39  0.396
#> Balanced   Balanced 3.46  0.439
#> Happy         Happy 3.55  0.550
sort_by(item_scores, ~ net)
#>                item mean    net
#> Angry         Angry 1.37 -0.936
#> Sad             Sad 1.70 -0.871
#> Afraid       Afraid 1.65 -0.864
#> Worried     Worried 2.19 -0.629
#> Exhausted Exhausted 2.66 -0.314
#> Lively       Lively 2.79 -0.168
#> Relaxed     Relaxed 3.30  0.304
#> Cheerful   Cheerful 3.39  0.396
#> Balanced   Balanced 3.46  0.439
#> Happy         Happy 3.55  0.550

The two orders agree except for one pair: by mean, Afraid (1.65) comes before Sad (1.70); by net score, Sad (−0.871) comes before Afraid (−0.864). The mean uses every level, the net score only whether a response lies above or below the midpoint; they diverge when two items have similar means but differ in how many responses sit exactly at the midpoint.

sequential_dist(ema_emotions, sort_by = "net")

The bands now run from Angry, the item with the lowest net score, to Happy, the item with the highest.

Reading order: flow and start_from

flow = "snake" (the default of the survey functions) alternates the direction of the bands, as in a boustrophedon; flow = "natural" reads every band in the same direction. start_from decides on which side the first band begins. The survey functions return their layout invisibly, and as.data.frame() gives its bands table, which records the geometry of each band: direction is the side the path enters from (which always alternates, because the arcs alternate), and read_direction is the direction in which the content of the band is read.

natural <- survey_sequence(ema_emotions,
                           colors = snake_palette("earth", n = 5),
                           flow = "natural", label_align = "left")

subset(as.data.frame(natural), select = c(i, direction, read_direction))
#>    i direction read_direction
#> 1  0       ltr            ltr
#> 2  1       rtl            ltr
#> 3  2       ltr            ltr
#> 4  3       rtl            ltr
#> 5  4       ltr            ltr
#> 6  5       rtl            ltr
#> 7  6       ltr            ltr
#> 8  7       rtl            ltr
#> 9  8       ltr            ltr
#> 10 9       rtl            ltr

With flow = "natural" the direction column alternates between ltr and rtl while read_direction is ltr for all ten bands.

mirrored <- survey_sequence(ema_emotions,
                            colors = snake_palette("earth", n = 5),
                            start_from = "right", label_align = "direction")

subset(as.data.frame(mirrored), select = c(i, direction, read_direction))
#>    i direction read_direction
#> 1  0       rtl            rtl
#> 2  1       ltr            ltr
#> 3  2       rtl            rtl
#> 4  3       ltr            ltr
#> 5  4       rtl            rtl
#> 6  5       ltr            ltr
#> 7  6       rtl            rtl
#> 8  7       ltr            ltr
#> 9  8       rtl            rtl
#> 10 9       ltr            ltr

With the default flow = "snake" and start_from = "right", the first band is read from right to left (rtl) and the reading direction alternates with the path. label_align = "direction" places each item label on the side where its band begins to be read, whereas "left" (the default) and "right" place all labels on one side. In survey plots each band is an item, and the segments within a band follow the order of the scale, not of time, so the response levels are drawn from left to right in every band; reverse_rtl = TRUE mirrors them on right-to-left bands. When the content of a band is ordered in time, as in sequence_snake() or activity_snake(), the reading direction changes what the reader sees; see the sequences guide.

Geometry and figure size

All geometry is in plot units: band_height is the height of a band, band_gap the white space between bands and plot_width the length of a band. The canvas grows with all three; as.data.frame() with what = "canvas" gives its size.

default_layout <- survey_sequence(ema_emotions,
                                  colors = snake_palette("earth", n = 5))

as.data.frame(default_layout, what = "canvas")
#>   width height
#> 1   692    506
compact_layout <- survey_sequence(ema_emotions,
                                  colors = snake_palette("earth", n = 5),
                                  band_height = 20, band_gap = 8,
                                  plot_width = 400)

as.data.frame(compact_layout, what = "canvas")
#>   width height
#> 1   570    372

The default layout of the ten items is 692 by 506 units, an aspect ratio of about 1.37; the compact layout is 570 by 372 units, about 1.53. The canvas is scaled to the graphics device, so the ratio of fig.width to fig.height should be close to the ratio of the canvas width to its height; otherwise white space appears on two sides of the plot. Thinner bands and narrower gaps are the way to keep a plot of many items from becoming tall: here they reduce the height by 134 units, about a quarter.

Summary

  • Pass n equal to the number of response levels to snake_palette(); the survey functions use only the first length(levels) colours.
  • Use a diverging palette with survey_snake() or survey_sequence() for bipolar scales with a neutral midpoint, and sequential_dist() or a sequential palette for intensity and frequency scales.
  • Order items with sort_by = "mean" or "net"; the two differ only where items have close means but different shares at the midpoint.
  • flow = "natural" reads every band in the same direction; flow = "snake" alternates; start_from selects the side of the first band; label_align = "direction" follows the reading direction.
  • Thinner bands (band_height), narrower gaps (band_gap) and shorter bands (plot_width) shrink the canvas; match the figure’s aspect ratio to it.

Reference

Neubauer, A. B., & Schmiedek, F. (2024). Approaching academic adjustment on multiple time scales. Zeitschrift für Erziehungswissenschaft, 27(1), 147–168. https://doi.org/10.1007/s11618-023-01182-8