"""Draw original Week 10 scenarios from exact, provenance-labeled lab CSVs. Run from the book directory after the lab smoke script: conda run -n research python assets/scripts/build-week10-research-design.py The transport values are instructor-known scenarios, not realized estimates. Missing-outcome ranges are identification ranges, not confidence intervals. """ from pathlib import Path import csv import math import os import sys BOOK = Path(__file__).resolve().parents[2] REPO = BOOK.parents[1] LAB = BOOK / 'labs/10-frontier-questions-personnel-data-and-student-research-designs' PROVENANCE = 'SYNTHETIC_TEACHING_CONSTRUCTION_NOT_PUBLISHED_REPLICATION' os.environ.setdefault('MPLCONFIGDIR', str(BOOK / '.mpl-cache')) import matplotlib matplotlib.use('Agg') import matplotlib.pyplot as plt sys.path.insert(0, str(REPO)) from shared.figure_style import COLORS, apply_style, save_figure, style_axis def read_and_preserve(name, target): source = LAB / 'output/diagnose' / name destination = BOOK / 'assets/data' / target destination.parent.mkdir(parents=True, exist_ok=True) destination.write_bytes(source.read_bytes()) with destination.open(newline='') as stream: rows = list(csv.DictReader(stream)) assert rows and all(row['provenance'] == PROVENANCE for row in rows) return rows def save(fig, stem): destination = BOOK / 'assets/figures' / stem save_figure(fig, destination.with_suffix('.png')) fig.savefig(destination.with_suffix('.pdf'), bbox_inches='tight', facecolor='white', metadata={'CreationDate': None, 'ModDate': None}) plt.close(fig) def main(): scenarios = read_and_preserve('transport_scenarios.csv', '10-transport-scenarios.csv') ranges = read_and_preserve('worker_followup_bounds.csv', '10-worker-followup-bounds.csv') assert [row['scenario'] for row in scenarios] == [ 'source_mix', 'target_stable', 'target_changed_implementation'] effects = [float(row['effect']) for row in scenarios] assert effects == [6, 3.5, .5] for row in scenarios: p = float(row['high_share']) assert math.isclose(float(row['effect']), p * float(row['high_effect']) + (1-p) * float(row['low_effect']), abs_tol=1e-8) apply_style() fig, ax = plt.subplots(figsize=(10.6, 5.5)) bars = ax.bar(range(3), effects, width=.55, color=[COLORS[c] for c in ['navy', 'teal', 'rust']]) ax.bar_label(bars, labels=[f'{value:g}' for value in effects], padding=6, fontsize=13) ax.set(title='Same target composition, different implementation', ylabel='Effect in invented productivity units', ylim=(0, 7.5), xticks=range(3), xticklabels=[ 'Source\nHigh-support share: 2/3\nConditional effects: 8, 2', 'Stable target\nHigh-support share: 1/4\nConditional effects: 8, 2', 'Changed implementation\nHigh-support share: 1/4\nConditional effects: 5, −1']) style_axis(ax) ax.axhline(effects[1], color=COLORS['teal'], linewidth=1, linestyle='--') ax.text(2, 3.8, 'Source standardization\nstill predicts 3.5', ha='center', color=COLORS['teal'], fontsize=10) fig.text(.5, .035, 'Original synthetic scenarios • instructor-known effects • no target outcomes observed', ha='center', fontsize=10, color=COLORS['muted']) fig.tight_layout(rect=(0, .10, 1, 1)) save(fig, '10-transport-and-implementation') by_object = {row['object']: row for row in ranges} assert set(by_object) == {'treated_full_mean', 'control_full_mean', 'treated_minus_control'} for name in ['treated_full_mean', 'control_full_mean']: row = by_object[name] p, mean = float(row['response_rate']), float(row['respondent_mean']) assert math.isclose(float(row['lower']), p * mean) assert math.isclose(float(row['upper']), p * mean + (1-p) * 100) treated, control = by_object['treated_full_mean'], by_object['control_full_mean'] difference = by_object['treated_minus_control'] assert float(difference['lower']) == float(treated['lower']) - float(control['upper']) assert float(difference['upper']) == float(treated['upper']) - float(control['lower']) fig, axes = plt.subplots(1, 2, figsize=(10.6, 5.5), gridspec_kw={'width_ratios': [1.15, 1]}) for y, row, label, color in [(1, treated, 'Treated', 'teal'), (0, control, 'Control', 'navy')]: lo, hi = float(row['lower']), float(row['upper']) axes[0].hlines(y, lo, hi, color=COLORS[color], linewidth=5) axes[0].plot([lo, hi], [y, y], '|', color=COLORS[color], markersize=15, markeredgewidth=2) axes[0].text((lo+hi)/2, y+.2, f'{label}: [{lo:g}, {hi:g}]', ha='center', fontsize=11) axes[0].text((lo+hi)/2, y-.25, f"Response {float(row['response_rate']):.0%}; respondent mean {float(row['respondent_mean']):g}", ha='center', fontsize=9) axes[0].set(title='Full-population outcome means', xlabel='Outcome on a 0–100 scale', xlim=(0, 100), ylim=(-.7, 1.7), yticks=[]) lo, hi = float(difference['lower']), float(difference['upper']) axes[1].hlines(0, lo, hi, color=COLORS['rust'], linewidth=5) axes[1].plot([lo, hi], [0, 0], '|', color=COLORS['rust'], markersize=15, markeredgewidth=2) axes[1].axvline(0, color=COLORS['muted'], linestyle='--', linewidth=1) axes[1].text((lo+hi)/2, .2, f'[{lo:g}, {hi:g}]', ha='center', fontsize=12) axes[1].text(0, -.35, 'Either sign remains possible', ha='center', fontsize=10) axes[1].set(title='Treated minus control', xlabel='Difference in outcome units', xlim=(-40, 40), ylim=(-.7, 1.7), yticks=[]) for ax in axes: style_axis(ax, grid_axis='x') fig.text(.5, .095, 'Arbitrary missingness + bounded outcomes • identification ranges, not confidence intervals or Lee bounds', ha='center', fontsize=9) fig.text(.5, .035, 'Original synthetic example • respondent difference of 10 does not determine the population sign', ha='center', fontsize=10, color=COLORS['muted']) fig.tight_layout(rect=(0, .15, 1, 1), w_pad=2.5) save(fig, '10-missing-outcome-ranges') print('PASS: two exact lab CSV inputs; scenario arithmetic and missing-outcome bounds checked; two PNG/PDF pairs saved.') if __name__ == '__main__': main()