aid-analysis 0.1.0

Abnormal Impact Differential (AID) analysis with ANOVA significance testing for year-over-year event impact measurement.

aid-analysis

Abnormal Impact Differential (AID) — a difference-in-differences style tool for measuring how much a metric deviated from its normal year-over-year pattern during a specific event window, with a two-way ANOVA significance test included.

Installation

pip install aid-analysis

What it does

Given daily (or other tidy) time-series data for one or more numeric metrics, AID asks:

"How much did my metric change from Period 1 to Period 2 this year, compared to how much it normally changes from Period 1 to Period 2 in a baseline year?"

The answer is a single number in percentage points (the AID), plus a p-value from a two-way ANOVA testing whether that difference is statistically significant.

The formula

$$\text{AID} = 100 \times \left[\left(\frac{\mu_{\text{current}, P_2}}{\mu_{\text{current}, P_1}} - 1\right) - \left(\frac{\mu_{\text{baseline}, P_2}}{\mu_{\text{baseline}, P_1}} - 1\right)\right]$$

A negative AID means the metric performed worse in the current year than prior-year seasonality would predict; positive means better.

Quick start

from aid_analysis import run_analysis, PeriodConfig

cfg = PeriodConfig(
    current_year=2026,
    baseline_year=2025,
    period_1=("01-29", "02-28"),   # "before" window, inclusive
    period_2=("03-01", "03-31"),   # "after" window, inclusive
)

# Single metric
summary = run_analysis("data.csv", metrics="Revenue", config=cfg)

# Multiple metrics
summary = run_analysis(
    "data.csv",
    metrics=["Transactions", "Count of Users", "Revenue"],
    config=cfg,
)

Your CSV just needs a date column and one or more numeric columns.

Input data format

A tidy CSV (or pandas.DataFrame) with:

• a date column (any parseable date format — you can change the column name via PeriodConfig(date_col="..."))
• one or more numeric metric columns (any names, including names with spaces like "Count of Users")

Example:

date,Transactions,Revenue
2025-01-29,142,8930.50
2025-01-30,156,9420.00
...
2026-03-31,98,6110.25

Output

Three things:

1. A printed report with the means, step-by-step AID calculation, the full ANOVA table, and plain-English interpretation.
2. A CSV summary (aid_summary.csv by default) with one row per metric.
3. A pandas DataFrame returned from run_analysis() for further work.

Example summary:

metric	mean_2026_period_1	mean_2026_period_2	mean_2025_period_1	mean_2025_period_2	change_2026_pct	change_2025_pct	aid_pp	anova_interaction_pvalue
Revenue	8,430.50	6,120.75	8,200.10	8,350.40	-27.39	1.83	-29.22	0.00023

Class-based API

For more control over intermediate steps:

from aid_analysis import AbnormalImpactAnalyzer, PeriodConfig

cfg = PeriodConfig(
    current_year=2026,
    baseline_year=2025,
    period_1=("01-29", "02-28"),
    period_2=("03-01", "03-31"),
)

analyzer = AbnormalImpactAnalyzer("data.csv", config=cfg)
analyzer.run(metrics=["Revenue", "Transactions"])
analyzer.print_report()

# Access individual results
revenue_result = analyzer.results["Revenue"]
print(revenue_result.delta_pp)            # the AID value
print(revenue_result.anova_table)         # full ANOVA table
print(revenue_result.anova_interaction_pvalue)

# Get the summary as a DataFrame
df = analyzer.summary_frame()

# Or export to CSV
analyzer.export("my_summary.csv")

How to interpret the results

AID value	p-value	Meaning
Large negative	< 0.05	Real, significant underperformance vs the seasonal baseline
Large negative	≥ 0.05	Looks like a drop, but could be random noise
Large positive	< 0.05	Real, significant outperformance vs the seasonal baseline
Large positive	≥ 0.05	Looks like an uplift, but could be random noise
Near zero	any	No abnormal impact vs the prior-year pattern

Requirements

• Python 3.9+
• pandas ≥ 1.5
• numpy ≥ 1.22
• statsmodels ≥ 0.13

License

MIT