| MarketMatching {MarketMatching} | R Documentation |
Market Matching and Causal Impact Inference
Description
For a given test market find the best matching control markets using time series matching and analyze the impact of an intervention (prospective or historical). The intervention could be be a marketing event or some other local business tactic that is being tested. The package utilizes dynamic time warping to do the matching and the CausalImpact package to analyze the causal impact. In fact, MarketMatching is simply a wrapper and workflow for those two packages. MarketMatching does not provide any functionality that cannot be found in these packages but simplifies the workflow of using dtw and CausalImpact together. In addition, if you don't already have a set of test markets to match, 'MarketMatching' can provide suggested test/control market pairs using the 'suggest_market_splits' option in the ‘best_matches()' function. Also, the 'test_fake_lift()' function provides pseudo prospective power analysis if you’re using the 'MarketMatching' package to create your test design (i.e., not just doing the post inference).
Details
The MarketMatching package can be used to perform the following analyses:
- For all markets in the input dataset, find the best control markets using time series matching.
- Given a test market and a matching control market (from above), analyze the causal impact of an intervention.
- Create optimal test/control market splits and run pseudo prospective power analyses.
The package utilizes the dtw package in CRAN to do the time series matching, and the CausalImpact package to do the inference. (Created by Kay Brodersen at Google). For more information about the CausualImpact package, see the following reference:
CausalImpact version 1.0.3, Brodersen et al., Annals of Applied Statistics (2015). http://google.github.io/CausalImpact/
The MarketMatching has two separate functions to perform the tasks described above:
- best_matches(): This function finds the best matching control markets for all markets in the input dataset. If you don't know the test markets the function can also provide suggested optimized test/control pairs.
- inference(): Given an object from best_matches(), this function analyzes the causal impact of an intervention.
- test_fake_lift(): Calculate the probability of a causal impact for fake interventions (prospective pseudo power).
For more details, check out the vignette: browseVignettes("MarketMatching")
Author(s)
Kim Larsen (kblarsen4 at gmail.com)
Examples
## Not run:
##-----------------------------------------------------------------------
## Find best matches for CPH
## If we leave test_market as NULL, best matches are found for all markets
##-----------------------------------------------------------------------
library(MarketMatching)
data(weather, package="MarketMatching")
mm <- MarketMatching::best_matches(data=weather,
id="Area",
date_variable="Date",
matching_variable="Mean_TemperatureF",
parallel=FALSE,
markets_to_be_matched="CPH",
warping_limit=1, # warping limit=1
dtw_emphasis=1, # rely only on dtw for pre-screening
matches=5, # request 5 matches
start_match_period="2014-01-01",
end_match_period="2014-10-01")
head(mm$Distances)
##-----------------------------------------------------------------------
## Analyze causal impact of a made-up weather intervention in Copenhagen
## Since this is weather data it is a not a very meaningful example.
## This is merely to demonstrate the functionality.
##-----------------------------------------------------------------------
results <- MarketMatching::inference(matched_markets = mm,
test_market = "CPH",
analyze_betas=FALSE,
end_post_period = "2015-10-01",
prior_level_sd = 0.002)
## Plot the impact
results$PlotCumulativeEffect
## Plot actual observations for test market (CPH) versus the expectation (based on the control)
results$PlotActualVersusExpected
##-----------------------------------------------------------------------
## Power analysis for a fake intervention ending at 2015-10-01
## The maximum lift analyzed is 5 percent, the minimum is 0 (using 5 steps)
## Since this is weather data it is a not a very meaningful example.
## This is merely to demonstrate the functionality.
##-----------------------------------------------------------------------
power <- MarketMatching::test_fake_lift(matched_markets = mm,
test_market = "CPH",
end_fake_post_period = "2015-10-01",
prior_level_sd = 0.002,
steps=20,
max_fake_lift=0.05)
## Plot the curve
power$ResultsGraph
##-----------------------------------------------------------------------
## Generate suggested test/control pairs
##-----------------------------------------------------------------------
data(weather, package="MarketMatching")
mm <- MarketMatching::best_matches(data=weather,
id_variable="Area",
date_variable="Date",
matching_variable="Mean_TemperatureF",
suggest_market_splits=TRUE,
parallel=FALSE,
dtw_emphasis=1, # rely only on correlation for this analysis
start_match_period="2014-01-01",
end_match_period="2014-10-01")
##-----------------------------------------------------------------------
## The file that contains the suggested test/control splits
## The file is sorted from the strongest market pair to the weakest pair.
##-----------------------------------------------------------------------
head(mm$SuggestedTestControlSplits)
##-----------------------------------------------------------------------
## Pass the results to test_fake_lift to get pseudo power curves for the splits.
## This tells us how well the design can detect various lifts.
## Not a meaningful example for this data. Just to illustrate.
## Note that the rollup() function will aggregate the test and control markets.
## The new aggregated test markets will be labeled "TEST."
##-----------------------------------------------------------------------
rollup <- MarketMatching::roll_up_optimal_pairs(matched_markets = mm,
synthetic=FALSE)
power <- MarketMatching::test_fake_lift(matched_markets = rollup,
test_market = "TEST",
end_fake_post_period = "2015-10-01",
lift_pattern_type = "constant",
max_fake_lift = 0.1)
## End(Not run)