Plotly error bars: error_y in px and graph_objects (tested)

In plotly express pass a column name: px.scatter(df, x="x", y="y", error_y="sd") (add error_y_minus= for asymmetric bars). In graph_objects pass a dict: go.Scatter(..., error_y=dict(type="data", array=sd)). Plotly does not aggregate, so compute the mean and sd per group first.

1. Every way to add error bars

b1.py prints the error_y object each call produces:

import pandas as pd, plotly, plotly.express as px, plotly.graph_objects as go
print("plotly", plotly.__version__)
df = pd.DataFrame({"x": ["a", "b", "c"], "y": [4.0, 6.0, 5.0],
                   "sd": [0.5, 1.0, 0.8], "lo": [0.2, 0.4, 0.3]})

fig = px.scatter(df, x="x", y="y", error_y="sd")                 # symmetric
print("px symmetric:", fig.data[0].error_y.to_plotly_json())

fig = px.bar(df, x="x", y="y", error_y="sd", error_y_minus="lo") # asymmetric
print("px asymmetric:", fig.data[0].error_y.to_plotly_json())

fig = go.Figure(go.Scatter(x=df.x, y=df.y, mode="markers",
                           error_y=dict(type="data", array=df.sd)))
print("go data:", fig.data[0].error_y.to_plotly_json())

fig = go.Figure(go.Bar(x=df.x, y=df.y, error_y=dict(type="percent", value=10)))
print("go percent:", fig.data[0].error_y.to_plotly_json())

try:
    px.scatter(df, x="x", y="y", error_y="std")
except ValueError as ex:
    print("ValueError:", str(ex).splitlines()[0])
try:
    px.scatter(df, x="x", y="y", error_y=[0.1, 0.2])
except ValueError as ex:
    print("ValueError:", str(ex).splitlines()[0])

Output (plotly 7.1.0, pandas 3.1.0, Python 3.13.5):

plotly 7.1.0
px symmetric: {'array': array([0.5, 1. , 0.8])}
px asymmetric: {'array': array([0.5, 1. , 0.8]), 'arrayminus': array([0.2, 0.4, 0.3])}
go data: {'array': array([0.5, 1. , 0.8]), 'type': 'data'}
go percent: {'type': 'percent', 'value': 10}
ValueError: Value of 'error_y' is not the name of a column in 'data_frame'. Expected one of ['x', 'y', 'sd', 'lo'] but received: std
ValueError: All arguments should have the same length. The length of argument `error_y` is 2, whereas the length of previously-processed arguments ['x', 'y'] is 3

• plotly express: error_y="sd" takes a column name (or an array of the same length as the data) and fills error_y.array. Add error_y_minus="lo" for asymmetric bars (arrayminus). error_x / error_x_minus work the same way horizontally.
• graph_objects: pass a dict, error_y=dict(type="data", array=...) for absolute values, or dict(type="percent", value=10) for ±10% of each y. For asymmetric bars add symmetric=False, arrayminus=....
• A misspelled column name gives Value of 'error_y' is not the name of a column, and a list of the wrong length gives All arguments should have the same length.

2. Error bars from raw rows: aggregate first

b2.py:

import pandas as pd, plotly.express as px
raw = pd.DataFrame({"group": ["a"] * 4 + ["b"] * 4,
                    "value": [3, 5, 4, 6, 7, 9, 8, 10]})
# px does not aggregate: compute mean and spread per group first
stats = raw.groupby("group", as_index=False)["value"].agg(mean="mean", sd="std", n="count")
stats["se"] = stats["sd"] / stats["n"] ** 0.5
print(stats.round(3).to_string(index=False))
fig = px.bar(stats, x="group", y="mean", error_y="sd")
print("bars:", fig.data[0].y.tolist(), "error_y:", fig.data[0].error_y.array.round(3).tolist())

fig = px.bar(raw, x="group", y="value")  # raw rows, no error bars
print("raw rows -> traces:", len(fig.data), "bar segments:", len(fig.data[0].y))

group  mean    sd  n    se
    a   4.5 1.291  4 0.645
    b   8.5 1.291  4 0.645
bars: [4.5, 8.5] error_y: [1.291, 1.291]
raw rows -> traces: 1 bar segments: 8

• Plotly express does not compute means or standard deviations for you: px.bar on raw rows draws one segment per row (8 here, stacked) and no error bars. Group with pandas, compute std (or the standard error sd / sqrt(n)), and pass that column to error_y.
• If you want the mean and a bootstrap confidence interval computed automatically, seaborn does that: see seaborn errorbar: sd vs ci vs se vs pi. For matplotlib, see matplotlib errorbar.

Written with AI assistance. Measured on 2026-10-23 with plotly 7.1.0, pandas 3.1.0 and Python 3.13.5 on Linux. Scripts ran verbatim and the output was pasted unedited. PlainTables is our extension; Plotly is free.