fix(impact): do not estimate impact with too little data

backend-python/media_impact_monitor/impact.py

@@ -33,6 +33,13 @@ def get_impact(q: ImpactSearch) -> Impact:
             end_date=q.end_date,
         )
     )
+    n_event_days = events["date"].nunique()
+    if n_event_days < 5:
+        return Impact(
+            method_applicability=False,
+            method_limitations=["Not enough events to estimate impact."],
+            impact_estimates=None,
+        )
     q.impacted_trend.start_date = q.start_date
     q.impacted_trend.end_date = q.end_date
     trends = get_trend(TrendSearch(**dict(q.impacted_trend)))
@@ -55,7 +62,7 @@ def get_impact(q: ImpactSearch) -> Impact:
     assert (
         len(set([str(lims) for lims in lims_list])) == 1
     ), "All topics should have same limitations."
-    n_days = 14 - 1
+    n_days = 7 - 1
     return Impact(
         method_applicability=applicabilities[0],
         method_limitations=lims_list[0],

backend-python/media_impact_monitor/impact_estimators/time_series_regression.py

@@ -20,7 +20,7 @@ def add_emws(df: pd.DataFrame, spans=[1, 2, 7, 30, 90, 365]):
     """Add new columns with exponentially weighted moving averages."""
     emws = pd.DataFrame(
         {
-            f"{col}_emw{i}": df.shift(1).ewm(span=i).mean().iloc[:, 0]
+            f"{col}_emw{i}": df.shift(1).ewm(halflife=i).mean().iloc[:, 0]
             for i in spans
             for col in df.columns
         }
@@ -46,10 +46,10 @@ def regress(
     """Get regression result where the outcome is `day` days after the treatment."""
     lags = range(1, lags + 1)
     media_df = pd.DataFrame(media_df, columns=["count"])
-    protest_df = add_lags(protest_df, lags=lags)
-    media_df = add_lags(media_df, lags=lags)
+    # protest_df = add_lags(protest_df, lags=[])
+    media_df = add_lags(media_df, lags=[4,5,6,7,8])
     # protest_df = add_emws(protest_df)
-    media_df = add_emws(media_df, spans=[7, 30, 90])
+    # media_df = add_emws(media_df, spans=[14])
     df = pd.concat([protest_df, media_df], axis=1)
     df = add_weekday_dummies(df)
     treatment = "protest"
@@ -63,16 +63,20 @@ def regress(
         else:
             df[outcome] = df[outcome].rolling(day + 1).sum()
     df = df.dropna()
+    placebo = False
+    if placebo:
+        df[treatment] = df.sample(frac=1)[treatment].to_list()
     X = df.drop(columns=[outcome])
     y = df[outcome]
     model = sm.OLS(y, sm.add_constant(X))
     model = model.fit(cov_type="HC3")
+    alpha = 0.1
     return {
         "date": day,
         "mean": model.params[treatment],
         "p": model.pvalues[treatment],
-        "ci_lower": model.conf_int()[0][treatment],
-        "ci_upper": model.conf_int()[1][treatment],
+        "ci_lower": model.conf_int(alpha=alpha)[0][treatment],
+        "ci_upper": model.conf_int(alpha=alpha)[1][treatment],
     }
 
 

frontend-observable/src/impacts.md

@@ -12,7 +12,8 @@ let impact = await queryApi('impact', {
     media_source: 'news_print',
     topic: 'climate_change'
   },
-  organizer: 'Last Generation (Germany)',
+  organizer: 'Fridays for Future',
+  start_date: '2020-04-10',
   end_date: '2022-04-30'
 })
 display(impact)