Grouped Bar Charts: Compare Two Metrics Side by Side

The entire trick is half a bar's width. Nudge two bar series apart by that much and you've got a grouped bar — which answers a different question than the stacked bar does. Stacked asks "what is this total made of?"; grouped asks "within each team, which is bigger — A or B?" When you want to set a team's offense beside its defense rather than pile one on the other, that half-width offset is the move. The example below is a grouped offense-vs-defense chart for the NBA's best teams.

This builds on Your First Visualization. The data is the bundled nba_ratings.csv (real 2023-24 NBA team ratings), so it runs offline.

1. The data: two ratings per team

   Each team has an offensive rating (points scored per 100 possessions) and a defensive rating (points allowed per 100). We take the ten best teams by net rating and shorten the names to nicknames for tidy axis labels.

   import pandas as pd
   
   df = pd.read_csv("nba_ratings.csv")
   df["Short"] = df["Team"].str.split().str[-1]   # "Boston Celtics" -> "Celtics"
   top = df.sort_values("NRtg", ascending=False).head(10).reset_index(drop=True)
   print(top[["Short", "ORtg", "DRtg", "NRtg"]].to_string())

   Offense and defense for the top 10

             Short    ORtg    DRtg   NRtg
   0       Celtics  124.23  112.51  11.71
   1       Thunder  120.43  113.12   7.31
   2  Timberwolves  116.63  109.98   6.65
   3       Nuggets  119.49  114.05   5.44
   4        Knicks  119.11  114.22   4.89
   5      Pelicans  118.31  113.70   4.61
   6      Clippers  119.75  116.51   3.24
   7          Suns  118.78  115.69   3.09
   8         76ers  117.79  114.76   3.02
   9        Pacers  121.85  118.85   2.99

   Notice the ratings cluster between roughly 108 and 125 — a narrow band where the differences that matter are only a few points. That will shape how we set up the axis.

2. The half-width offset

   Make one set of evenly spaced x positions with np.arange, then draw the first series shifted left by half a bar and the second shifted right by half a bar. They share each team's slot without overlapping.

   import numpy as np
   import matplotlib.pyplot as plt
   
   x = np.arange(len(top))   # 0, 1, 2, ... one slot per team
   width = 0.4
   
   fig, ax = plt.subplots(figsize=(9, 6))
   ax.bar(x - width/2, top["ORtg"], width, label="Offensive rating", color="#C56A1E")
   ax.bar(x + width/2, top["DRtg"], width, label="Defensive rating", color="#2C5E8A")
   ax.set_xticks(x)
   ax.set_xticklabels(top["Short"], rotation=45, ha="right")
   ax.set_ylim(100, 128)     # zoom to the band the ratings actually occupy
   ax.legend()
   fig.savefig("grouped_bars.png", dpi=144, bbox_inches="tight")

   

   Setting the x-ticks to the bare x positions (not the offset ones) centers each label under its pair. With more than two series, the pattern generalizes: for n series, offset bar i by (i − (n−1)/2) × width.

3. The zoomed axis: useful, but flag it

   We set ylim(100, 128) so a three-point rating gap is visible instead of being a rounding error against a bar that starts at zero. That's a legitimate choice for comparing values in a narrow band — but it's also the classic way a chart can exaggerate small differences. The honest move is to make the truncation obvious: keep the axis labeled clearly, and when a difference looks dramatic, remember the bars don't start at zero. For an audience that might over-read the gaps, plot net rating directly instead, which is naturally centered.

Troubleshooting