Causal Impact
Last updated
Last updated
With Causal Impact Analysis, you can estimate impact of an event (advertisement, etc.) on a market, by using other markets to form a synthetic baseline.
Input data should be a time series data frame. It should have following columns.
Date/Time Column - A date/time column with Date or POSIXct data.
Measure Column - A numeric column with the values that you want to estimate the impact of the event upon.
Market Column - A character or factor column that has ID/name of the market for the row.
You can access this feature from 'Add' (Plus) button. Mouse over "Run Analytics ..." menu, and select "Run Market Impact Analysis".
You can also access this feature from the column menu of the Measure Column. Mouse over "Run Analytics ..." menu and select "Run Market Impact Analysis" menu.
Select date/time column with "Date / Time Column" dropdown.
Select aggregation level from "Aggregation Level" dropdown. Input data is aggregated with this time unit. For example, if you select "Day" and there are multiple rows that falls within a same day, the values for those rows are aggregated to form single row for the date. As a result, this becomes the time unit for the resulting time series data frame. Aggregation level can be one of the following.
Day
Week
Month
Quarter
Year
Select a numeric column from "Measure Column" dropdown. This is the value of interest that you want to estimate the impact of the event upon.
Select aggregation function from "Aggregation Function" dropdown. This is how the multiple rows that falls under same time period (the one you specified with "Aggregation Level".) is aggregated. Its value can be one of the following.
sum
mean
count
median
max
min
first
last
mad - Median Absolute Deviation
sd - Standard Deviation
var - Variance
Select a character/factor column from "Market Column" dropdown that has IDs/names of markets.
Select a value from Market Column that is the market of interest.
In the "Event Time" field, specify the date when the event of interest happened.
Specify optional parameters in "Parameters" section.
Number of Predictor Markets
NA Fill Type
Output Type
Weight on Distance (vs Correlation) for Ranking Markets
Tail-Area Probability for Posterior Interval
You can control the confidence level of the confidence intervals in the output. The default is 0.05 which translates into 95% confidence level.
Number of MCMC Samples
Standardize Data
Prior Standard Deviation of Random Walk
Period of Seasonal Trend
Duration of a Season
Dynamic Regression
Click "Run" button.
The output data is time series with original measure, synthetic baseline, and estimated impact of the event upon the market of interest. Following is the list of output columns.
time - Time for each row.
actual - Actual values from the target market.
expected - Expected values of the target market if it were not for the event of the interest (advertisement etc.). Synthetic control calculated from values of other markets.
expected_high / expected_low - Confidence interval of expected.
impact - Impact of the event calculated as the difference between actual value and expected value.
impact_high / impact_low - Confidence interval of impact.
cumulative_impact - Cumulated sum of the impact.
cumulative_impact_high / cumulative_impact_low - Confidence interval of cumulative_impact.
actual_at_event_time - Actual value at the event time. Useful to mark the event time at visualization.
