Making coefficient plots in Python using Forestplot

Lucas Shen — Wed, 28 Sep 2022 12:45:27 +0000

This post illustrates how one can use the open-source Forestplot package to plot estimates with confidence intervals.

This package plots correlation coefficients or regression estimates from upstream analyses (see this example of correlation analysis).

Prepare the package and load the data

To install the package from PyPI:



pip install forestplot

Load example dataset that reports how certain factors correlate with the amount of sleep one gets:



import forestplot as fp

df = fp.load_data("sleep")  # companion example data
df.head(3)

	var	r	moerror	label	group	ll	hl	n	power	p-val
0	age	0.0903729	0.0696271	in years	age	0.02	0.16	706	0.67	0.0163089
1	black	-0.0270573	0.0770573	=1 if black	other factors	-0.1	0.05	706	0.11	0.472889
2	clerical	0.0480811	0.0719189	=1 if clerical worker	occupation	-0.03	0.12	706	0.25	0.201948

In the above dataframe, each row is an individual characteristic with a corresponding correlation coefficient from correlating the characteristic with the amount of sleep one gets per night.

The first row, age, for instance, with a correlation coefficient of 0.09 (p = 0.016), says that people who are older get more sleep.
(See this notebook to see how the correlation coefficients are computed from the real sleep75.csv data.)

Plot the estimates

Forest plots (or coefficient plots, dot plots, coefplots) are useful to visualize the estimates and their confidence intervals.

To plot the estimates in df:



fp.forestplot(df,  # the dataframe with results data
              estimate="r",  # col containing estimated effect size 
              ll="ll", hl="hl",  # columns containing conf. int. lower and higher limits
              varlabel="label",  # column containing variable label
              ylabel="Confidence interval",  # y-label title
              xlabel="Pearson correlation"  # x-label title
              )

Customizing and adding annotations (Pt. 1)

You can add variable group subheadings (e.g. the Labor Factors subheading) and sort the estimates (within groups). You can also sort the order of the variable group subheadings (group_order):



fp.forestplot(df,  # the dataframe with results data
              estimate="r",  # col containing estimated effect size 
              moerror="moerror",  # columns containing conf. int. margin of error
              varlabel="label",  # column containing variable label
              groupvar="group",  # Add variable groupings 
              # group ordering
              group_order=["labor factors", "occupation", "age", "health factors", 
                           "family factors", "area of residence", "other factors"],
              sort=True  # sort in ascending order (sorts within group if group is specified)               
              )

Customizing and adding annotations (Pt. 2)

You can also add more annotations to the plot, such as the sample size (e.g. N and formatted_pval) and add table lines:



fp.forestplot(df,  # the dataframe with results data
              estimate="r",  # col containing estimated effect size 
              ll="ll", hl="hl",  # lower & higher limits of conf. int.
              varlabel="label",  # column containing the varlabels to be printed on far left
              pval="p-val",  # column containing p-values to be formatted
              annote=["n", "power", "est_ci"],  # columns to report on left of plot
              annoteheaders=["N", "Power", "Est. (95% Conf. Int.)"],  # ^corresponding headers
              rightannote=["formatted_pval", "group"],  # columns to report on right of plot 
              right_annoteheaders=["P-value", "Variable group"],  # ^corresponding headers
              groupvar="group",  # column containing group labels
              group_order=["labor factors", "occupation", "age", "health factors", 
                           "family factors", "area of residence", "other factors"],                   
              xlabel="Pearson correlation coefficient",  # x-label title
              xticks=[-.4,-.2,0, .2],  # x-ticks to be printed
              sort=True,  # sort estimates in ascending order
              table=True,  # Format as a table
              # Additional kwargs for customizations
              **{"marker": "D",  # set maker symbol as diamond
                 "markersize": 35,  # adjust marker size
                 "xlinestyle": (0, (10, 5)),  # long dash for x-reference line 
                 "xlinecolor": ".1",  # gray color for x-reference line
                 "xtick_size": 12,  # adjust x-ticker fontsize
                }  
              )

Final remarks

Planned future enhancements include allowing for multiple estimates per row in the plot.

Forest plots have many aliases. Other names include coefplots, coefficient plots, meta-analysis plots, dot plots, dot-and-whisker plots, blobbograms, margins plots, regression plots, and ropeladder plots.

This posts hopefully gives my forestplot package some visibility. At the the same time, happy to hear comments about the API's ease of use and features. plot. See the GitHub repo readme for a more substantial documentation.

LSYS / forestplot

A Python package to make publication-ready but customizable coefficient plots.

Forestplot

Easy API for forest plots.
A Python package to make publication-ready but customizable forest plots

This package makes publication-ready forest plots easy to make out-of-the-box. Users provide a dataframe (e.g. from a spreadsheet) where rows correspond to a variable/study with columns including estimates, variable labels, and lower and upper confidence interval limits. Additional options allow easy addition of columns in the dataframe as annotations in the plot.

Release
Status
Coverage
Python
Docs
Meta
Binder

show/hide

Installation

Quick Start

Some Examples with Customizations

Gallery and API Options

Multi-models

Known Issues

Background and Additional Resources

Contributing

Installation

Install from PyPI

pip install forestplot

Install from conda-forge

conda install forestplot

Install from source

git clone https://github.com/LSYS/forestplot.git
cd forestplot
pip install .

Developer installation

git clone https://github.com/LSYS/forestplot.git
cd forestplot
pip install -r requirements_dev.txt

make lint
make test

(back to top)

Quick Start

import forestplot as fp
df =

…

View on GitHub

DEV Community: Lucas Shen