8 Scatterplots

為了督促同學演練撰寫程式碼，移除原電子書的解答。請同學儘可能完成每項演練之前，不要查閱原電子書內容。各項解答將在課堂演練活動中說明。

In this chapter we will work with our data to generate a plot of two variables from the Woodworth et al. dataset. Before we get to generate our plot, we still need to work through the steps to get the data in the shape we need it to be in for our particular question. In particular we need to generate the object summarydata that just has the variable we need.You have done these steps before so go back to the relevant Lab and use that to guide you through.

8.0.1 Activity 1: Set-up

  • Open R Studio and ensure the environment is clear.
  • Open the stub-scatterplot.Rmd file and ensure that the working directory is set to your Data Skills folder and that the two .csv data files (participant-info.csv and ahi-cesd.csv) are in your working directory (you should see them in the file pane).
  • Look through your previous work to find the code that loads the tidyverse, reads in the data files and creates an object called all_dat that joins the two objects dat and pinfo.

8.0.2 Activity 2: Select

Select the columns ahiTotal, cesdTotal, sex, age, educ from the all_dat and save them in a new object named variable summarydata.

8.0.3 Activity 3: Simple scatterplots

First, we want to look at whether there seems to be a relationship between happiness and depression scores across all participants.

In order to visualise two continuous variables, we can use a scatterplot. Using the ggplot code you learned about in Intro to Data Viz, try and recreate the below plot.

A few hints:

  • Use the summarydata data.
  • Put ahiTotal on the x-axis and cesdTotal on the y-axis.
  • Rather than using geom_bar(), geom_violin(), or geom_boxplot(), for a scatterplot you need to use geom_point().
  • Rather than using scale_fill_viridis_d() to change the colour, add the argument colour = "red" to geom_point (except replace “red” with whatever colour you’d prefer).
  • Remember to edit the axis names.
Scatterplot of happiness and depression scores

Figure 8.1: Scatterplot of happiness and depression scores


How would you describe the relationship between the two variables?

8.0.4 Activity 4: Adding a line of best fit

Scatterplots are very useful but it can often help to add a line of best fit to help interpretation. Add the below layer to the scatterplot code from Activity 4:

  • This code uses the function geom_smooth() to draw the line. There are several different methods but we want a straight, or linear, line so we specify method = "lm".
  • This line is really a regression line, which you’ll learn more about in level 2. For now, the steeper the slope of the line, the stronger the relationship between the two variables.
  • By default the regression line will be extended, beyond the original y-axis limits, if you want to change this so that your plots looks like the below, add limits = c(0,60) to scale_y_continuous()
geom_smooth(method = "lm")
Scatterplot of happiness and depression scores

Figure 8.2: Scatterplot of happiness and depression scores

8.0.5 Activity 5: Setting the factors

It seems fairly obvious that there might be a negative relationship between happiness and depression, so instead we want to look at whether this relationship changes depending on different demographic variables.

Just like in the last chapter, we need to ensure that R knows what type of data each variable is. At the moment, sex, educ, and income are all registered as numeric variables, but we know that they’re really categories.

str(summarydata)
## tibble [992 x 6] (S3: tbl_df/tbl/data.frame)
##  $ ahiTotal : num [1:992] 32 34 34 35 36 37 38 38 38 38 ...
##  $ cesdTotal: num [1:992] 50 49 47 41 36 35 50 55 47 39 ...
##  $ sex      : num [1:992] 1 1 1 1 1 1 2 1 2 2 ...
##  $ age      : num [1:992] 46 37 37 19 40 49 42 57 41 41 ...
##  $ educ     : num [1:992] 4 3 3 2 5 4 4 4 4 4 ...
##  $ income   : num [1:992] 3 2 2 1 2 1 1 2 1 1 ...
  • Using the same method you used in the last chapter, overwrite sex, educ and income in summarydata as factors.

8.0.6 Activity 6: Grouped scatterplots

We can now use our factors to display the data in the scatterplots for each group.

  • Rather than adding colour to geom_point() which sets the colour for all the data points, instead we add colour = sex to the aesthetic mapping on the first line. This tells ggplot() to produce different colours for each level (or group) in the variable sex.
  • scale_color_viridis_d() works exactly like the other colour blind friendly scale functions you have used, so we can use name and labels to adjust the legend.
ggplot(summarydata, aes(x = ahiTotal , y = cesdTotal, colour = sex)) + 
  geom_point() +
  scale_x_continuous(name = "Happiness Score") +
  scale_y_continuous(name = "Depression Score",
                     limits = c(0,60)) +
  theme_minimal() +
  geom_smooth(method = "lm") +
  scale_color_viridis_d(name = "Participant sex", 
                       labels = c("Male", "Female"),
                       option = "E")

It looks like the relationship between happiness and depression is about the same for male and female participants.

  • Create two more scatterplots that show the relationship between happiness and depression grouped by a) income and b) educ. Make sure you update the legend labels (you might need to check the code book).

8.0.7 Activity 7: Group by a new variable

So, let’s be honest, there’s not much going on with any of the demographic variables - the relationship between depression and anxiety is pretty much the same for all of the groups. A reasonable hypothesis might be that rather than being connected to any demographic variables, the relationship between happiness and depression changes depending upon your general happiness level.

  • Using mutate create a new variable named happiness in summarydata that evaluates whether a participant’s happiness score is equal to or higher than the median ahiTotal score.
  • This is not an easy task as it’s not something we’ve explicitly shown you how to do but it’s the last activity for this semester so before you go to the solution, do a bit of trial and error, then look at the hints to see if you can get near the answer.
  • If you’ve done it right, summarydata should contain a column named happiness with the value TRUE if ahiTotal is above the overall median and FALSE if it is below.
mutate(data, new_variable = equal_or_more_than_median)
mutate(data, new_variable = variable >= median(variable))

Now, reproduce the below plot using this new variable:

ggplot(summarydata, aes(x = ahiTotal , y = cesdTotal, colour = happiness)) + 
  geom_point() +
  scale_x_continuous(name = "Happiness Score") +
  scale_y_continuous(name = "Depression Score",
                     limits = c(0,60)) +
  theme_minimal() +
  geom_smooth(method = "lm") +
  scale_colour_viridis_d(
    name = "Happiness",
    labels = c("Below median happiness",
                         "Above median happiness"))

What might you conclude from this plot?

The line for the “unhappy” group is much steeper than the line for the “happy” group. That is, is you’re generally unhappy, then you report a much stronger link between your happiness and depression levels than if you’re generally happy.

If you actually wanted to look at the difference in the relationships statistically, you could compute the correlation coefficients like below, but don’t worry about that until Level 2.

dat_above <- filter(summarydata, happiness == TRUE)
dat_below <- filter(summarydata, happiness == FALSE)

cor.test(dat_above$ahiTotal, dat_above$cesdTotal)
cor.test(dat_below$ahiTotal, dat_below$cesdTotal)
## 
##  Pearson's product-moment correlation
## 
## data:  dat_above$ahiTotal and dat_above$cesdTotal
## t = -7.6564, df = 491, p-value = 1.024e-13
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  -0.4032637 -0.2453473
## sample estimates:
##        cor 
## -0.3265828 
## 
## 
##  Pearson's product-moment correlation
## 
## data:  dat_below$ahiTotal and dat_below$cesdTotal
## t = -22.474, df = 497, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  -0.7509362 -0.6635268
## sample estimates:
##        cor 
## -0.7099551

8.0.7.1 Finished!

Great job! As you may have noticed, this chapter tried to push you and test what you’ve learned - we hope you can see just how far you’ve come in the space of just a couple of months, it’s genuinely amazing what you have achieved and you should feel proud. In Psych 1B we will continue using these wrangling skills on new data and also data that you collect yourself.