Lecture 4.2 KEY

Note

This template follows lecture 4.2 slides. Please be sure to cross-reference the slides, which contain important information and additional context!

Setup

##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
##                                    setup                                 ----
##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

#..........................load packages.........................
library(metajam) 
library(tidyverse)

#...................download data from DataOne...................
download_d1_data("https://cn.dataone.org/cn/v2/resolve/https%3A%2F%2Fpasta.lternet.edu%2Fpackage%2Fdata%2Feml%2Fknb-lter-hbr%2F208%2F11%2F3b3cf7ea447cb875d7c7d68ebdfd24c7",
                 path = here::here("week4")) 

#  ~ NOTE: You should rename the downloaded folder to 'data/' so that it's ignored by .gitignore! ~

#....................read in downloaded files....................
stream_chem_all <- read_d1_files(here::here("week4", "data"))

#........................get the data file.......................
stream_chem_data <- stream_chem_all$data

Scatter plots

Basic scatter plot

# basic scatter plot ----
p1 <- stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH)) + 
  geom_point(alpha = 0.5)

p1

Alt 1: add a rug plot

# scatter + rug plot ----
p1 +
  geom_rug()

Alt 2: marginal plot alternatives using {ggExtra}

# add marginal plots with {ggExtra} ----
# also try "density", "boxplot"
ggExtra::ggMarginal(p1, type = "histogram")

Alt 3: scatter + marginal plot with 2+ groups

# scatter plot with points colored by site ----
p2 <- stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH, color = site)) + 
  geom_point(alpha = 0.5) +
  theme(legend.position = "bottom")

# add marginal density plot ----
ggExtra::ggMarginal(p2, type = "density", groupFill = TRUE, groupColour = TRUE)

Trend lines

Default: LOESS method

• similar to a moving average

# add a smoothed line ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  filter(site == "W8") |> 
  ggplot(aes(x = DOC, y = pH)) + 
  geom_point(alpha = 0.5) +
  geom_smooth()

`geom_smooth()` using method = 'loess' and formula = 'y ~ x'

Line of best fit

# add a trend line ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  filter(site == "W8") |> 
  ggplot(aes(x = DOC, y = pH)) + 
  geom_point(alpha = 0.5) +
  geom_smooth(method = "lm", se = FALSE)

`geom_smooth()` using formula = 'y ~ x'

Visualizing a third numeric variable

Bubble charts

• variation of a scatter plot for visualizing a third numeric variable
• be extra mindful:
  • primary focus = relationship between your x- and y-axis variables
  • hard to compare the strengths of different associations (is there a better alternative graphic form?)
  • if the range of values mapped to size is small, your bubbles will look indistinguishable from one another
  • consider adjusting the size range of your scale

# bubble chart ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH, color = site, size = Al_ICP)) + 
  geom_point(alpha = 0.5) +
  labs(x = "DOC (mg/L)", size = "Al (mg/L)", color = "Site")

• adjust size range of bubbles

# scale size by area ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH, color = site, size = Al_ICP)) + 
  geom_point(alpha = 0.5) +
  scale_size(range = c(1, 10)) +
  labs(x = "DOC (mg/L)", size = "Al (mg/L)", color = "Site")

• IMPORTANT: don’t scale size by radius (can be deceptive)

# don't scale size by radius ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH, color = site, size = Al_ICP)) + 
  geom_point(alpha = 0.5) +
  scale_radius(range = c(1, 10)) +
  labs(x = "DOC (mg/L)", size = "Al (mg/L)", color = "Site")

Use color for third variable

stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH, color = Al_ICP)) + 
  geom_point(alpha = 0.5, size = 2) +
  scale_color_viridis_c() +
  labs(x = "DOC (mg/L)", color = "Al (mg/L)")

But oftentimes best to just create two different plots

# effect of DOC on pH ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = DOC, y = pH, color = site)) + 
  geom_point(alpha = 0.5)
# effect of DOC on Al_ICP ----
stream_chem_data |> 
  filter(waterYr == 2021) |> 
  ggplot(aes(x = Al_ICP, y = pH, color = site)) + 
  geom_point(alpha = 0.5)

Overplotting

• if you have too many points, scatter plots can be ineffective

Initial strategies

• adjust size / transparency of points

• add a rug plot (or marginal plot)

• color by group

Alt 1: heatmaps

• e.g. like you’re looking down on a histogram

• or use hexagonal shapes
• also increase height of legend can make continuous scale easier to read

Alt 2: 2d density / contour plots

• e.g. like you’re looking down on a density plot
• legend provides an estimate of the proportion of data points fall within a colored region (density of distribution of points sums to 1)
  • interpreting legend: 0-2% of of points fall within a 1x1 square in the darkest blue region, while 26-28% fall within a 1x1 square in the brightest yellow region

Alt 3: 2d density plots using {ggdensity}

• can be easier to interpret
• compute and plot the resulting highest density regions (HDRs)
• HDRs are computed to be the smallest such regions that bound that level of probability