create.boxplot: how to add multiple lines of math expressions?

[jarbet]

For create.boxplot, I am trying to add 2 lines of math expressions: a p-value and Spearman correlation (since the group variable is ordinal).

However, when I pass a vector of 2 expression objects to text.labels, it does not display both elements but instead displays the first element twice:

library(survival);
library(BoutrosLab.plotting.general);
#> Loading required package: lattice
#> Loading required package: latticeExtra
#> Loading required package: cluster
#> Loading required package: hexbin
#> Loading required package: grid
#> 
#> Attaching package: 'BoutrosLab.plotting.general'
#> The following object is masked from 'package:stats':
#> 
#>     dist

data(cancer);
lung$ph.ecog.factor <- factor(lung$ph.ecog);

# correlation
spearman <- cor.test(
    x = as.numeric(lung$age),
    y = lung$ph.ecog,
    method = 'spearman',
    use = 'complete.obs'
    );
#> Warning in cor.test.default(x = as.numeric(lung$age), y = lung$ph.ecog, : Cannot
#> compute exact p-value with ties
spearman.p <- spearman$p.value;
spearman.rho <- round(spearman$estimate, 2);
names(spearman.rho) <- NULL;
text.pvalue <- display.statistical.result(
    x = spearman.p,
    statistic.type = 'p',
    symbol = ' = '
    );
text.rho <-  as.expression(bquote(rho == .(spearman.rho)));

create.boxplot(
    formula = age ~ ph.ecog.factor,
    data = lung,
    add.stripplot = TRUE,
    add.text = TRUE,
    text.labels = c(text.rho, text.pvalue),
    text.x = c(4, 4),
    text.y = c(max(lung$age, na.rm = TRUE), max(lung$age, na.rm = TRUE) - 3),
    text.fontface = 'plain'
    )
#> Warning in validDetails.text(x): NAs introduced by coercion

^{Created on 2023-06-22 by the reprex package (v2.0.1)}

[stefaneng]

You can use atop within bquote:

text.pvalue <- display.statistical.result(
    x = spearman.p,
    statistic.type = 'p',
    symbol = ' = '
    );

text.label <- as.expression(
  bquote(atop(rho == .(spearman.rho), .(as.character(text.pvalue))))
)

create.boxplot(
    formula = age ~ ph.ecog.factor,
    data = lung,
    add.stripplot = TRUE,
    add.text = TRUE,
    text.labels = text.label,
    text.x = 4,
    text.y = max(lung$age, na.rm = TRUE) - 1,
    text.fontface = 'plain'
    )

[jarbet]

You can use atop within bquote:

I noticed you used as.character for the second expression, which works in this example. But what if the p-value is really small and you need to use an expression for scientific notation?

For example:

text.pvalue <- display.statistical.result(
    x = 0.00000000512,
    statistic.type = 'p',
    symbol = ' = '
    );

How could you combine that with atop such that the p-value appears like:

[stefaneng]

You can use atop within bquote:

I noticed you used as.character for the second expression, which works in this example. But what if the p-value is really small and you need to use an expression for scientific notation?

For example:

text.pvalue <- display.statistical.result(
    x = 0.00000000512,
    statistic.type = 'p',
    symbol = ' = '
    );

How could you combine that with atop such that the p-value appears like:

Got it:

text.pvalue <- display.statistical.result(
    x = 0.00000000512,
    statistic.type = 'p',
    symbol = ' = '
    );

text.label <- as.expression(
  bquote(atop(rho == .(spearman.rho[[1]]), .(text.pvalue[[1]])))
)

create.boxplot(
    formula = age ~ ph.ecog.factor,
    data = lung,
    add.stripplot = TRUE,
    add.text = TRUE,
    text.labels = text.label,
    text.x = 4,
    text.y = max(lung$age, na.rm = TRUE) - 1,
    text.fontface = 'plain'
    )

Source: https://stackoverflow.com/a/59074689

[pboutros]

I might be misunderstanding, but take a look at how it's done for create.km.plot in BPS, my recollection is that the canonical solution is to create a legend and that text.labels was explicitly designed for simple cases. You'll rarely want it for anything complex I suspect.

[pboutros]

How could you combine that with atop such that the p-value appears like:

Got it:

Yup I think this is entirely avoidable, shouldn't need atop or bquote for this per se. I think this is in the category of "could be done, but there are easier ways to do it". May be worth looking at other published code and verifying they use legend rather than text.labels as well.

[jarbet]

How could you combine that with atop such that the p-value appears like:

Got it:

Yup I think this is entirely avoidable, shouldn't need atop or bquote for this per se. I think this is in the category of "could be done, but there are easier ways to do it". May be worth looking at other published code and verifying they use legend rather than text.labels as well.

I checked create.km.plot and it uses a legend.

Another example is BPG::get.corr.key which creates an expression object of length 5 that is passed to the key argument of BPG::create.scatterplot.

I see create.boxplot also has a key argument, so I should be able to modify the code from get.corr.key if I want more than 2 lines. Otherwise @stefaneng's solution works for 2 lines.

[jarbet]

Here's an example using key for an expression vector of length 3:

library(survival);
library(BoutrosLab.plotting.general);
#> Loading required package: lattice
#> Loading required package: latticeExtra
#> Loading required package: cluster
#> Loading required package: hexbin
#> Loading required package: grid
#> 
#> Attaching package: 'BoutrosLab.plotting.general'
#> The following object is masked from 'package:stats':
#> 
#>     dist

data(cancer);
lung$ph.ecog.factor <- factor(lung$ph.ecog);

# correlation
spearman <- cor.test(
    x = as.numeric(lung$age),
    y = lung$ph.ecog,
    method = 'spearman',
    use = 'complete.obs'
    );
#> Warning in cor.test.default(x = as.numeric(lung$age), y = lung$ph.ecog, : Cannot
#> compute exact p-value with ties

spearman.p <- spearman$p.value;
spearman.rho <- round(spearman$estimate, 2);
names(spearman.rho) <- NULL;

text.pvalue <- display.statistical.result(
    x = 0.00000000512,
    statistic.type = 'p',
    symbol = ' = '
    );
text.rho <- as.expression(bquote(rho == .(spearman.rho[[1]])));
text.label <- as.expression(
    bquote(atop(rho == .(spearman.rho[[1]]), .(text.pvalue[[1]])))
    );

key.list <- c(text.rho, text.pvalue, expression(gamma))
key <- list(
    text = list(
        lab = key.list, 
        cex = 1
        ), 
    x = 0.7,
    y = 1
    );

create.boxplot(
    formula = age ~ ph.ecog.factor,
    data = lung,
    add.stripplot = TRUE,
    key = key
    )

^{Created on 2023-06-22 by the reprex package (v2.0.1)}

[stefaneng]

@jarbet Since there doesn't seem to be an actual issue and you have a good solution now this is probably best as a discussion