Seaborn Statistical Estimation & Data Handling

🎨 The core semantic mappings

sns.scatterplot(data=df, x="x", y="y", hue="category", size="value", style="group")

Param	Maps a variable to	Works with
hue=	color	most plot types
size=	point/line size or width	scatter, line
style=	marker shape / line dashes	scatter, line

hue + style together on the same variable = redundant encoding (color AND shape both show the category) — genuinely helpful for colorblind accessibility or grayscale printing.

📏 Confidence intervals / error bars

sns.lineplot(data=df, x="x", y="y", errorbar="sd")             # standard deviation
sns.lineplot(data=df, x="x", y="y", errorbar=("ci", 95))          # confidence interval, 95%
sns.lineplot(data=df, x="x", y="y", errorbar=("pi", 50))             # percentile interval
sns.lineplot(data=df, x="x", y="y", errorbar=None)                      # no error band at all
sns.barplot(data=df, x="cat", y="y", errorbar="se")                        # standard error

Seaborn computes these automatically via bootstrapping when there are multiple y-values per x. This is the main thing that separates Seaborn's lineplot/barplot from plain matplotlib equivalents.

🧮 estimator — how repeated values get aggregated

sns.barplot(data=df, x="day", y="total_bill", estimator="mean")      # default
sns.barplot(data=df, x="day", y="total_bill", estimator="median")
sns.barplot(data=df, x="day", y="total_bill", estimator="sum")
sns.barplot(data=df, x="day", y="total_bill", estimator=len)            # count

📐 Long vs wide format data

# WIDE format — one column per category
#    Male  Female
# 0   23      25
# 1   19      22
 
# LONG format — one row per observation, category as a value
#    sex     value
# 0  Male     23
# 1  Female   25

wide_to_long = df.melt(id_vars="id", var_name="sex", value_name="value")   # pandas: wide → long

Seaborn generally expects LONG format ( x=, y=, hue= as column names). Most real-world CSVs come wide — df.melt() is the standard fix.

sns.boxplot(data=long_df, x="sex", y="value")      # works cleanly on long data

🧾 Handling missing data

sns.scatterplot(data=df.dropna(subset=["x", "y"]), x="x", y="y")   # drop NaNs before plotting

Seaborn generally silently drops NaN rows for the relevant columns rather than erroring — good to know when a plot has fewer points than expected.

🔢 Ordering categories explicitly

sns.boxplot(data=df, x="day", y="total_bill", order=["Thur", "Fri", "Sat", "Sun"])
sns.barplot(data=df, x="day", y="total_bill", hue="sex", hue_order=["Male", "Female"])

Without order=, category axis order is whatever pandas/Seaborn infers (often alphabetical) — explicitly set it whenever a natural order (days, sizes, ratings) matters.

🔗 Related

Seaborn Categorical Plots · Seaborn Relational Plots · Seaborn