sns.lineplot produces confusing output

[jtmbeta]

pipr.csv ss.csv

The following code produces lineplots with garbled confidence intervals and average traces despite no obvious issues with the data. Seaborn version 0.13.2, matplotlib 3.8.3.

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

pipr = pd.read_csv('pipr.csv')
ss = pd.read_csv('ss.csv')

fig, axs = plt.subplots(1, 2, figsize=(12,4),layout='tight')
sns.lineplot(
    data=pipr,
    x="time",
    y="pc_pupil",
    hue='condition',
    errorbar='ci',
    palette={'red': 'tab:red', 'blue': 'tab:blue'},
    ax=axs[0]
)

sns.lineplot(
    data=ss,
    x="time",
    y="pc_pupil",
    hue='condition',
    errorbar='ci',
    palette={'lms': 'tab:green', 'mel': 'tab:blue'},
    ax=axs[1]
)
for ax in axs:
    ax.set(xlabel="Time (s)", ylabel="Pupil size (%-change)")
    ax.fill_between(
        (0, 3), min(ax.get_ylim()), max(ax.get_ylim()), alpha=0.2, color="k"
    )
    ax.grid()
![Figure_2](https://github.com/user-attachments/assets/b0eafe42-d886-4c6c-be77-db9af65106f1)

[mwaskom]

Hi, not being familiar with your data, it's not obvious to me what is "wrong" here, or why it might be a seaborn problem.

[jtmbeta]

If I change the code to:

fig, axs = plt.subplots(1, 2, figsize=(12,4),layout='tight')
sns.lineplot(
    data=pipr,
    x="time",
    y="pc_pupil",
    hue='condition',
    errorbar='se',
    palette={'red': 'tab:red', 'blue': 'tab:blue'},
    ax=axs[0],
    units='subject',
    estimator=None
)

sns.lineplot(
    data=ss,
    x="time",
    y="pc_pupil",
    hue='condition',
    errorbar='se',
    palette={'lms': 'tab:green', 'mel': 'tab:blue'},
    ax=axs[1],
    units='subject',
    estimator=None
)
for ax in axs:
    ax.set(xlabel="Time (s)", ylabel="Pupil size (%-change)")
    ax.fill_between(
        (0, 3), min(ax.get_ylim()), max(ax.get_ylim()), alpha=0.2, color="k"
    )
    ax.grid()

We can now see the underlying traces in the data. I would expect a smooth trace for the average and shaded regions for the standard error. But the result is clearly wrong in the previous figure!

[mwaskom]

Your subjects do not have identical time values:

pipr.query("condition == 'blue'").pivot(index="time", columns="subject", values="pc_pupil")

subject           3005        3011        3012
time                                          
-1.000000   100.882803   99.998503  101.361294
-0.979978          NaN  100.059489  101.272379
-0.979978   100.883722         NaN         NaN
-0.959956          NaN  100.116923  101.190397
-0.959956   100.880034         NaN         NaN
...                ...         ...         ...
 16.959956         NaN   95.398760   90.926791
 16.959956   88.411231         NaN         NaN
 16.979978   88.535902         NaN         NaN
 16.979978         NaN   95.449669   91.013283
 17.000000   88.667938   95.502635   91.100839

sns.lineplot only aggregates across identical values of the x variable. If your data need some binning / interpolation, you'll have to do it yourself as a preprocessing step.

[jtmbeta]

Ahh, got it. Thank you!