passive_imputation.png

def _figure_passive(naive_pairs: pd.DataFrame, passive_pairs: pd.DataFrame) -> None:
    apply_style()
    fig, axes = plt.subplots(1, 2, figsize=(11, 6.2), sharex=True, sharey=True)

    all_vals = pd.concat([
        naive_pairs[["bmi", "bmi_von_gewicht"]],
        passive_pairs[["bmi", "bmi_von_gewicht"]],
    ])
    lo, hi = float(all_vals.min().min()), float(all_vals.max().max())
    pad = 0.04 * (hi - lo)
    lo, hi = lo - pad, hi + pad

    panels = [
        ("Naiv: bmi und gewicht_kg\nunabhängig imputiert", naive_pairs, PRIMARY),
        ("Passiv: bmi aus imputiertem\ngewicht_kg abgeleitet", passive_pairs, "#5B9E6E"),
    ]
    for ax, (title, data, color) in zip(axes, panels):
        ax.plot([lo, hi], [lo, hi], color=SECONDARY, linestyle="--", linewidth=1.3,
                label="Identität", zorder=1)
        ax.scatter(data["bmi_von_gewicht"], data["bmi"], s=20, alpha=0.5,
                   color=color, edgecolor="none", zorder=2)
        ax.set_title(title, fontsize=12)
        ax.set_xlabel("berechneter BMI aus imputiertem Gewicht (kg/m²)")
        ax.set_xlim(lo, hi)
        ax.set_ylim(lo, hi)
        ax.set_aspect("equal", adjustable="box")

    axes[0].set_ylabel("imputierter BMI (kg/m²)")
    axes[0].legend(loc="upper left", fontsize=9.5)
    fig.suptitle("Passive Imputation vermeidet die BMI-Inkonsistenz",
                 fontsize=14, fontweight="bold", x=0.01, y=1.0, ha="left")
    fig.tight_layout(rect=(0.0, 0.0, 1.0, 0.90))
    save(fig, ASSETS / "passive_imputation.png")