Skip to content

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

License

Notifications You must be signed in to change notification settings

LSYS/forestplot

Repository files navigation

Forestplot

PyPI - Python Version
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 PyPI Conda (channel only) GitHub release (latest by date)
Status CI Notebooks
Coverage Codecov
Python PyPI - Python Version
Docs Read the Docs (version) DocLinks
Meta GitHub Imports: isort Code style: black types - Mypy DOI
Binder Binder

Table of Contents

show/hide

Installation

Install from PyPI
PyPI

pip install forestplot

Install from conda-forge
Conda (channel only)

conda install forestplot

Install from source
GitHub release (latest by date)

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 = 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 0.671578 0.0163089
1 black -0.0270573 0.0770573 =1 if black other factors -0.1 0.05 706 0.110805 0.472889
2 clerical 0.0480811 0.0719189 =1 if clerical worker occupation -0.03 0.12 706 0.247768 0.201948

(* This is a toy example of how certain factors correlate with the amount of sleep one gets. See the notebook that generates the data.)

The example input dataframe above have 4 key columns
Column Description Required
var Variable label
r Correlation coefficients (estimates to plot)
label Variable labels
group Variable grouping labels
ll Conf. int. lower limits
hl Containing the conf. int. higher limits
n Sample size
power Statistical power
p-val P-value

(See Gallery and API Options for more details on required and optional arguments.)

Make the forest plot

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
              )

Save the plot

plt.savefig("plot.png", bbox_inches="tight")

(back to top)

Some Examples With Customizations

  1. Add variable groupings, add group order, and sort by estimate size.
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
              capitalize="capitalize",  # Capitalize labels
              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)               
              )

  1. Add p-values on the right and color alternate rows gray
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
              capitalize="capitalize",  # Capitalize labels
              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)               
              pval="p-val",  # Column of p-value to be reported on right
              color_alt_rows=True,  # Gray alternate rows
              ylabel="Est.(95% Conf. Int.)",  # ylabel to print
              **{"ylabel1_size": 11}  # control size of printed ylabel
              )

  1. Customize annotations and make it a table
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
              capitalize="capitalize",  # Capitalize labels
              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
              xlabel="Pearson correlation coefficient",  # x-label title
              table=True,  # Format as a table
              )

  1. Strip down all bells and whistle
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
              capitalize="capitalize",  # Capitalize labels
              ci_report=False,  # Turn off conf. int. reporting
              flush=False,  # Turn off left-flush of text
              **{'fontfamily': 'sans-serif'}  # revert to sans-serif                              
              )

  1. Example with more customizations
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
              capitalize="capitalize",  # Capitalize labels
              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": "#808080",  # gray color for x-reference line
                 "xtick_size": 12,  # adjust x-ticker fontsize
                }  
              )

Annotations arguments allowed include:
  • ci_range: Confidence interval range (e.g. (-0.39 to -0.25)).
  • est_ci: Estimate and CI (e.g. -0.32(-0.39 to -0.25)).
  • formatted_pval: Formatted p-values (e.g. 0.01**).

To confirm what processed columns are available as annotations, you can do:

processed_df, ax = fp.forestplot(df, 
                                 ...  # other arguments here
                                 return_df=True  # return processed dataframe with processed columns
                                )
processed_df.head(3)
label group n r CI95% p-val BF10 power var hl ll moerror formatted_r formatted_ll formatted_hl ci_range est_ci formatted_pval formatted_n formatted_power formatted_est_ci yticklabel formatted_formatted_pval formatted_group yticklabel2
0 Mins worked per week Labor factors 706 -0.321384 [-0.39 -0.25] 1.99409e-18 1.961e+15 1 totwrk -0.25 -0.39 0.0686165 -0.32 -0.39 -0.25 (-0.39 to -0.25) -0.32(-0.39 to -0.25) 0.0*** 706 1 -0.32(-0.39 to -0.25) Mins worked per week 706 1.0 -0.32(-0.39 to -0.25) 0.0*** Labor factors 0.0*** Labor factors
1 Years of schooling Labor factors 706 -0.0950039 [-0.17 -0.02] 0.0115515 1.137 0.72 educ -0.02 -0.17 0.0749961 -0.1 -0.17 -0.02 (-0.17 to -0.02) -0.10(-0.17 to -0.02) 0.01** 706 0.72 -0.10(-0.17 to -0.02) Years of schooling 706 0.72 -0.10(-0.17 to -0.02) 0.01** Labor factors 0.01** Labor factors

(back to top)

Multi-models

For coefficient plots where each variable can have multiple estimates (each model has one).

import forestplot as fp

df_mmodel = pd.read_csv("../examples/data/sleep-mmodel.csv").query(
    "model=='all' | model=='young kids'"
)
df_mmodel.head(3)
var coef se T pval r2 adj_r2 ll hl model group label
0 age 0.994889 1.96925 0.505213 0.613625 0.127289 0.103656 -2.87382 4.8636 all age in years
3 age 22.634 15.4953 1.4607 0.149315 0.178147 -0.0136188 -8.36124 53.6293 young kids age in years
4 black -84.7966 82.1501 -1.03222 0.302454 0.127289 0.103656 -246.186 76.5925 all other factors =1 if black
fp.mforestplot(
    dataframe=df_mmodel,
    estimate="coef",
    ll="ll",
    hl="hl",
    varlabel="label",
    capitalize="capitalize",
    model_col="model",
    color_alt_rows=True,
    groupvar="group",
    table=True,
    rightannote=["var", "group"],
    right_annoteheaders=["Source", "Group"],
    xlabel="Coefficient (95% CI)",
    modellabels=["Have young kids", "Full sample"],
    xticks=[-1200, -600, 0, 600],
    mcolor=["#CC6677", "#4477AA"],
    # Additional kwargs for customizations
    **{
        "markersize": 30,
        # override default vertical offset between models (0.0 to 1.0)
        "offset": 0.35,  
        "xlinestyle": (0, (10, 5)),  # long dash for x-reference line
        "xlinecolor": ".8",  # gray color for x-reference line
    },
)

Please note: This module is still experimental. See this jupyter notebook for more examples and tweaks.

Gallery and API Options

Notebooks

Check out this jupyter notebook for a gallery variations of forest plots possible out-of-the-box. The table below shows the list of arguments users can pass in. More fined-grained control for base plot options (eg font sizes, marker colors) can be inferred from the example notebook gallery.

Option Description Required
dataframe Pandas dataframe where rows are variables (or studies for meta-analyses) and columns include estimated effect sizes, labels, and confidence intervals, etc.
estimate Name of column in dataframe containing the estimates.
varlabel Name of column in dataframe containing the variable labels (study labels if meta-analyses).
ll Name of column in dataframe containing the conf. int. lower limits.
hl Name of column in dataframe containing the conf. int. higher limits.
logscale If True, make the x-axis log scale. Default is False.
capitalize How to capitalize strings. Default is None. One of "capitalize", "title", "lower", "upper", "swapcase".
form_ci_report If True (default), report the estimates and confidence interval beside the variable labels.
ci_report If True (default), format the confidence interval as a string.
groupvar Name of column in dataframe containing the variable grouping labels.
group_order List of group labels indicating the order of groups to report in the plot.
annote List of columns to add as annotations on the left-hand side of the plot.
annoteheaders List of column headers for the left-hand side annotations.
rightannote List of columns to add as annotations on the right-hand side of the plot.
right_annoteheaders List of column headers for the right-hand side annotations.
pval Name of column in dataframe containing the p-values.
starpval If True (default), format p-values with stars indicating statistical significance.
sort If True, sort variables by estimate values in ascending order.
sortby Name of column to sort by. Default is estimate.
flush If True (default), left-flush variable labels and annotations.
decimal_precision Number of decimal places to print. (Default = 2)
figsize Tuple indicating core figure size. Default is (4, 8)
xticks List of xticklabels to print on x-axis.
ylabel Y-label title.
xlabel X-label title.
color_alt_rows If True, shade out alternating rows in gray.
preprocess If True (default), preprocess the dataframe before plotting.
return_df If True, returned the preprocessed dataframe.

(back to top)

Known Issues

  • Variable labels coinciding with group variables may lead to unexpected formatting issues in the graph.
  • Left-flushing of annotations relies on the monospace font.
  • Plot may give strange behavior for few rows of data (six rows or fewer. see this issue)
  • Plot can get cluttered with too many variables/rows (~30 onwards)
  • Not tested with PyCharm (#80) nor Google Colab (#110).
  • Duplicated varlabel may lead to unexpected results (see #76, #81). mplot for grouped models could be useful for such cases (see #59, WIP).

(back to top)

Background and Additional Resources

More about forest plots

Forest plots have many aliases (h/t Chris Alexiuk). Other names include coefplots, coefficient plots, meta-analysis plots, dot-and-whisker plots, blobbograms, margins plots, regression plots, and ropeladder plots.

Forest plots in the medical and health sciences literature are plots that report results from different studies as a meta-analysis. Markers are centered on the estimated effect and horizontal lines running through each marker depicts the confidence intervals.

The simplest version of a forest plot has two columns: one for the variables/studies, and the second for the estimated coefficients and confidence intervals. This layout is similar to coefficient plots (coefplots) and is thus useful for more than meta-analyses.

More resources about forest plots

  • [1] Chang, Y., Phillips, M.R., Guymer, R.H. et al. The 5 min meta-analysis: understanding how to read and interpret a forest plot. Eye 36, 673–675 (2022).
  • [2] Lewis S, Clarke M. Forest plots: trying to see the wood and the trees BMJ 2001; 322 :1479

More about this package

The package is lightweight, built on pandas, numpy, and matplotlib.

It is slightly opinioniated in that the aesthetics of the plot inherits some of my sensibilities about what makes a nice figure. You can however easily override most defaults for the look of the graph. This is possible via **kwargs in the forestplot API (see Gallery and API options) and the matplotlib API.

Planned enhancements include forest plots where each row can have multiple coefficients (e.g. from multiple models).

Related packages

  • [1] [Stata] Jann, Ben (2014). Plotting regression coefficients and other estimates. The Stata Journal 14(4): 708-737.
  • [2] [Python] Meta-Analysis in statsmodels
  • [3] [Python] Matt Bracher-Smith's Forestplot
  • [4] [R] Solt, Frederick and Hu, Yue (2021) dotwhisker: Dot-and-Whisker Plots of Regression Results
  • [5] [R] Bounthavong, Mark (2021) Forest plots. RPubs by RStudio

(back to top)

Contributing

Contributions are welcome, and they are greatly appreciated!

Potential ways to contribute:

  • Raise issues/bugs/questions
  • Write tests for missing coverage
  • Add features (see examples notebook for a survey of existing features)
  • Add example datasets with companion graphs
  • Add your graphs with companion code

Issues

Please submit bugs, questions, or issues you encounter to the GitHub Issue Tracker. For bugs, please provide a minimal reproducible example demonstrating the problem (it may help me troubleshoot if I have a version of your data).

Pull Requests

Please feel free to open an issue on the Issue Tracker if you'd like to discuss potential contributions via PRs.

(back to top)