Analysing survey data with complex sampling in R

This section covers data visualisations using survey design-adjusted plots.

Load data

Remember that we “saved” our wrangled dataset to the working directory in the previous section as an RData file in our workshop folder.

Code

root <- "C://Users//s1769862//OneDrive - University of Edinburgh//SLTF-workshop-August2024//" # change with where the folder lies in your filepath
cws <- readRDS(file= file.path(root, "Data", "cws.RData"))

Code

design1 <- survey::svydesign(ids = ~SchoolRef, strata =~Strata, weights = ~Weight, data =cws, check.strata=T)

Intro to plots

We want to take advantage of R’s good graphics but the typical plotting options, e.g. base R plot or ggplot2, are not coded to intake a survey design object to make their calculations. This means that weighted estimates and standard error visualisations will be incorrect.

Therefore, we need to specify the sampling design in the plotting arguments.

Using only the survey package

The survey package has inbuilt plotting functions that work out-the-box.

Note

Include ~1 to specify the lack of a second variable in univariate plots.

Code

survey::svyhist(FeelPositiveFuture~1,design1)

Looking at the range, the distribution of positive feeling is right-skewed.

The syntax is very similar for a boxplot.

We can also perform some basic modification, including adding a main title with main = "title" .

Code

survey::svyboxplot(FeelPositiveFuture~1,design1,
                   main = "Feelings of positivity for the future, 12 year olds in England")

Multivariate plots

To investigate bivariate relationships with a boxplot we can replace the ~1 with a categorical variable.

Code

survey::svyboxplot(FeelPositiveFuture~HomeType,design1,
                   main = "Feelings of positivity by living situation, 12 year olds in England")

It appears that there is greater variance in feelings of optimism for those in a type of home other than a family or fostering situation.

Code

survey::svyplot(FeelPositiveFuture~SatisfiedHealth,
                design1,
                xlab = "Satisfaction with health",
                ylab = "Feelings of positivity",
                main = "Relationship between health satisfaction and positivity",
                style="grayhex")

Task 6

Now, try out your own svy plotting options!

Key functions for main plot types include:

svyhist
svybox
svyplot

Advanced: try to create both univariate and multivariate plots!

Code

## write your own code!

## tip: copy and past from previous code chunks!

Better plots

The survey plots are great, but they are pretty basic. I also find the syntax non-intuitive.

The ggsurvey function offers an excellent range of options for plotting with complex survey design. It is part of the larger questionr package, which packages several survey functions.

ggsurvey is (to my understanding!) akin to wrapper for ggplot2 that incorporates the estimate corrections provided by survey package calculations. Therefore, if you already know ggplot grammar, you know how to make survey design corrected plot ¹.

The hegemony of ggplot grammar also makes it much easier to get help debugging your plots on forums like Stack Overflow.

Code

ggsurvey(design1) +
    aes(x = FrequencyHelpHousework, fill = Gender) +
    geom_bar(position = "fill") +
  labs(title = "Frequency of housework by gender",
       subtitle = "Children in England, aged 12",
       caption = "Data drawn from age 12 of the Children's Worlds Survey")

As with other ggplot2 objects, we can make this a lot prettier by customising the colour schemes and themes.

We can see what colour schemes are available from RColorBrewer using

Code

RColorBrewer::display.brewer.all()

Code

ggsurvey(design1) +
    aes(x = FrequencyHelpHousework, fill = Gender) +
    geom_bar(position = "fill") + scale_fill_brewer(palette = "Blues") +
  labs(title = "Frequency of housework by gender",
       subtitle = "Children in England, aged 12",
       caption = "Data drawn from age 12 of the Children's Worlds Survey")

We can also utilise the ggthemes package for pre-set colour schemes and other plotting options.

Code

ggsurvey(design1) +
    aes(x = FrequencyHelpHousework, fill = Gender) +
    geom_bar(position = "fill") + scale_fill_brewer(palette = "Blues") +
  labs(title = "Frequency of housework by gender",
       subtitle = "Children in England, aged 12",
       caption = "Data drawn from age 12 of the Children's Worlds Survey") +
  ggthemes::theme_economist_white()

Changing geometries

There are many geometries available, including boxplots.

Important

The boxplot geometry requires an underlying package quantreg which is not currently included in the original install. If your are getting an error check to ensure that this package is installed and loaded in your R session.

Code

ggsurvey(design1) +
    aes(y=FeelPositiveFuture,
        x=HomeType,
        fill=HomeType) + geom_boxplot() +
  scale_fill_brewer(palette = "Paired") + theme_classic() +
  labs(title = "Feelings of positivity by living situation",
       subtitle = "Children in England, aged 12",
       caption = "Data drawn from age 12 of the Children's Worlds Survey")

Task 7

Make your own plot!

Try customising colours, labels, or types!

Look at the data documentation for questionr (the ggsurvey function is located on page 18) and ggplot2 and more ggplot2, including cheat sheets

Code

## write your own code! Look at the previous chunks for help.

Next: Go to Part 2.1: Regressions

Footnotes

Double-check the data documentation for plots with more complicated statistics, e.g. geom smooth functions.↩︎

--- title: "1.3: Survey adjusted plots" format: html: code-fold: true code-tools: true editor: markdown: wrap: 100 --- This section covers data visualisations using survey design-adjusted plots. ---------------------------------------------------------------------------------------------------- ```{r library load 2, echo=F, message=F, warning=F, include=F} library(survey) # handling survey data with complex sampling design library(haven) # for reading in non-native data formats (e.g. Stata's .dta files) library(labelled) # working with labelled data library(dplyr) # data manipulation library(ggplot2) # data visualisation library(questionr) # data visualisation with survey design objects library(gtsummary) # nice publication ready summary tables library(ggthemes) # change theme of ggplot objects library(RColorBrewer) # for colour schemes library(hexbin) # assist with plotting graphics library(quantreg) # supports boxplot functions for ggsurvey ``` ## Load data Remember that we "saved" our wrangled dataset to the working directory in the previous section as an `RData` file in our workshop folder. ```{r load cws rdata} root <- "C://Users//s1769862//OneDrive - University of Edinburgh//SLTF-workshop-August2024//" # change with where the folder lies in your filepath cws <- readRDS(file= file.path(root, "Data", "cws.RData")) ``` ```{r set survey design object} design1 <- survey::svydesign(ids = ~SchoolRef, strata =~Strata, weights = ~Weight, data =cws, check.strata=T) ``` ## Intro to plots We want to take advantage of R's good graphics but the typical plotting options, e.g. `base R` `plot` or `ggplot2`, are not coded to intake a survey design object to make their calculations. This means that weighted estimates and standard error visualisations will be incorrect. Therefore, we need to specify the sampling design in the plotting arguments. ## Using only the survey package The `survey` package has inbuilt plotting functions that work out-the-box. ::: callout-note Include `~1` to specify the lack of a second variable in univariate plots. ::: ```{r survey histogram} survey::svyhist(FeelPositiveFuture~1,design1) ``` Looking at the range, the distribution of positive feeling is right-skewed. The syntax is very similar for a boxplot. We can also perform some basic modification, including adding a main title with `main = "title"` . ```{r survey box plot} survey::svyboxplot(FeelPositiveFuture~1,design1, main = "Feelings of positivity for the future, 12 year olds in England") ``` ### Multivariate plots To investigate bivariate relationships with a boxplot we can replace the `~1` with a categorical variable. ```{r bivariate boxplot} survey::svyboxplot(FeelPositiveFuture~HomeType,design1, main = "Feelings of positivity by living situation, 12 year olds in England") ``` It appears that there is greater variance in feelings of optimism for those in a type of home other than a family or fostering situation. ```{r multivariate interval plot} survey::svyplot(FeelPositiveFuture~SatisfiedHealth, design1, xlab = "Satisfaction with health", ylab = "Feelings of positivity", main = "Relationship between health satisfaction and positivity", style="grayhex") ``` ## Task 6 Now, try out your own `svy` plotting options! Key functions for main plot types include: - `svyhist` - `svybox` - `svyplot` > Advanced: try to create both univariate and multivariate plots! ```{r task 6} ## write your own code! ## tip: copy and past from previous code chunks! ``` ## Better plots The survey plots are great, but they are pretty basic. I also find the syntax non-intuitive. The [ggsurvey](https://cran.r-project.org/web/packages/ggsurvey/ggsurvey.pdf) function offers an excellent range of options for plotting with complex survey design. It is part of the larger [questionr](https://juba.github.io/questionr/) package, which packages several survey functions. `ggsurvey` is (to my understanding!) akin to wrapper for `ggplot2` that incorporates the estimate corrections provided by `survey` package calculations. Therefore, if you already know ggplot grammar, you know how to make survey design corrected plot [^1]. [^1]: Double-check the data documentation for plots with more complicated statistics, e.g. geom smooth functions. The hegemony of ggplot grammar also makes it much easier to get help debugging your plots on forums like Stack Overflow. ```{r ggsurvey plot 1} ggsurvey(design1) + aes(x = FrequencyHelpHousework, fill = Gender) + geom_bar(position = "fill") + labs(title = "Frequency of housework by gender", subtitle = "Children in England, aged 12", caption = "Data drawn from age 12 of the Children's Worlds Survey") ``` As with other `ggplot2` objects, we can make this a lot prettier by customising the colour schemes and themes. We can see what colour schemes are available from `RColorBrewer` using ```{r colour schemes} RColorBrewer::display.brewer.all() ``` ```{r prettier plot} ggsurvey(design1) + aes(x = FrequencyHelpHousework, fill = Gender) + geom_bar(position = "fill") + scale_fill_brewer(palette = "Blues") + labs(title = "Frequency of housework by gender", subtitle = "Children in England, aged 12", caption = "Data drawn from age 12 of the Children's Worlds Survey") ``` We can also utilise the `ggthemes` package for pre-set colour schemes and other plotting options. ```{r prettier plot 2} ggsurvey(design1) + aes(x = FrequencyHelpHousework, fill = Gender) + geom_bar(position = "fill") + scale_fill_brewer(palette = "Blues") + labs(title = "Frequency of housework by gender", subtitle = "Children in England, aged 12", caption = "Data drawn from age 12 of the Children's Worlds Survey") + ggthemes::theme_economist_white() ``` ### Changing geometries There are many geometries available, including boxplots. ::: callout-important The boxplot geometry requires an underlying package `quantreg` which is not currently included in the original install. If your are getting an error check to ensure that this package is installed and loaded in your R session. ::: ```{r prettier boxplot, warning=F} ggsurvey(design1) + aes(y=FeelPositiveFuture, x=HomeType, fill=HomeType) + geom_boxplot() + scale_fill_brewer(palette = "Paired") + theme_classic() + labs(title = "Feelings of positivity by living situation", subtitle = "Children in England, aged 12", caption = "Data drawn from age 12 of the Children's Worlds Survey") ``` ---------------------------------------------------------------------------------------------------- ## Task 7 > Make your own plot! - Try customising colours, labels, or types! Look at the data documentation for [questionr](https://cran.r-project.org/web/packages/questionr/questionr.pdf) (the ggsurvey function is located on page 18) and [ggplot2](https://ggplot2.tidyverse.org/reference/) and [more ggplot2, including cheat sheets](https://ggplot2.tidyverse.org/) ```{r task 3} ## write your own code! Look at the previous chunks for help. ``` ::: {.callout-note appearance="minimal" icon="false"} **Next**: Go to Part 2.1: Regressions :::