tidysynth
is a tidy implementation the synthetic control
method in R
.
A synthetic control offers a way of evaluating the effect of an
intervention in comparative case studies. The method aims to model a
counterfactual unit using a weighted average of units that did not
receive the intervention. The effect of the intervention can be
estimated by comparing differences in the observed and synthetic time
series. See Abadie et al. 2003, 2010, 2015 for more on the method and
use cases.
Building on the Synth
package, tidysynth
makes a number of
improvements when implementing the method in R
. These improvements
allow users to inspect, visualize, and tune the synthetic control more
easily. A key benefit of a tidy implementation is that the entire
preparation process for building the synthetic control can be
accomplished in a single pipe.
Specifically, the package:
grab_
prefix functions
to easily extract component elements from synthetic control pipeline.Cran.
install.packages('tidysynth')
Developer Version.
# install.packages("devtools")
devtools::install_github("edunford/tidysynth")
The package uses a pipeline of functions to generate the synthetic control.
Function | Description |
---|---|
synthetic_control() |
Initialize a synth pipeline by specifying the panel series, outcome, and intervention period. This pipeline operates as a nested tbl_df |
generate_predictor() |
Create one or more scalar variables summarizing covariate data across a specified time window. These predictor variables are used to fit the synthetic control. |
generate_weights() |
Fit the unit and predictor weights used to generate the synthetic control. |
generate_control() |
Generate the synthetic control using the optimized weights. |
The following example comes from Abadie et al. 2010, which evaluates the impact of Proposition 99 on cigarette consumption in California.
require(tidysynth)
data("smoking")
smoking %>% dplyr::glimpse()
## Rows: 1,209
## Columns: 7
## $ state <chr> "Rhode Island", "Tennessee", "Indiana", "Nevada", "Louisiana…
## $ year <dbl> 1970, 1970, 1970, 1970, 1970, 1970, 1970, 1970, 1970, 1970, …
## $ cigsale <dbl> 123.9, 99.8, 134.6, 189.5, 115.9, 108.4, 265.7, 93.8, 100.3,…
## $ lnincome <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
## $ beer <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
## $ age15to24 <dbl> 0.1831579, 0.1780438, 0.1765159, 0.1615542, 0.1851852, 0.175…
## $ retprice <dbl> 39.3, 39.9, 30.6, 38.9, 34.3, 38.4, 31.4, 37.3, 36.7, 28.8, …
The method aims to generate a synthetic California using information from a subset of control states (the “donor pool”) where a similar law was not implemented. The donor pool is the subset of case comparisons from which information is borrowed to generate a synthetic version of the treated unit (“California”).
smoking_out <-
smoking %>%
# initial the synthetic control object
synthetic_control(outcome = cigsale, # outcome
unit = state, # unit index in the panel data
time = year, # time index in the panel data
i_unit = "California", # unit where the intervention occurred
i_time = 1988, # time period when the intervention occurred
generate_placebos=T # generate placebo synthetic controls (for inference)
) %>%
# Generate the aggregate predictors used to fit the weights
# average log income, retail price of cigarettes, and proportion of the
# population between 15 and 24 years of age from 1980 - 1988
generate_predictor(time_window = 1980:1988,
ln_income = mean(lnincome, na.rm = T),
ret_price = mean(retprice, na.rm = T),
youth = mean(age15to24, na.rm = T)) %>%
# average beer consumption in the donor pool from 1984 - 1988
generate_predictor(time_window = 1984:1988,
beer_sales = mean(beer, na.rm = T)) %>%
# Lagged cigarette sales
generate_predictor(time_window = 1975,
cigsale_1975 = cigsale) %>%
generate_predictor(time_window = 1980,
cigsale_1980 = cigsale) %>%
generate_predictor(time_window = 1988,
cigsale_1988 = cigsale) %>%
# Generate the fitted weights for the synthetic control
generate_weights(optimization_window = 1970:1988, # time to use in the optimization task
margin_ipop = .02,sigf_ipop = 7,bound_ipop = 6 # optimizer options
) %>%
# Generate the synthetic control
generate_control()
Once the synthetic control is generated, one can easily assess the fit by comparing the trends of the synthetic and observed time series. The idea is that the trends in the pre-intervention period should map closely onto one another.
smoking_out %>% plot_trends()
To capture the causal quantity (i.e. the difference between the observed
and counterfactual), one can plot the differences using
plot_differences()
smoking_out %>% plot_differences()
In addition, one can easily examine the weighting of the units and variables in the fit. This allows one to see which cases were used, in part, to generate the synthetic control.
smoking_out %>% plot_weights()
Another useful way of evaluating the synthetic control is to look at how comparable the synthetic control is to the observed covariates of the treated unit.
smoking_out %>% grab_balance_table()
## # A tibble: 7 × 4
## variable California synthetic_California donor_sample
## <chr> <dbl> <dbl> <dbl>
## 1 ln_income 10.1 9.85 9.83
## 2 ret_price 89.4 89.4 87.3
## 3 youth 0.174 0.174 0.173
## 4 beer_sales 24.3 24.2 23.7
## 5 cigsale_1975 127. 127. 137.
## 6 cigsale_1980 120. 120. 138.
## 7 cigsale_1988 90.1 91.4 114.
For inference, the method relies on repeating the method for every donor
in the donor pool exactly as was done for the treated unit —
i.e. generating placebo synthetic controls). By setting
generate_placebos = TRUE
when initializing the synth pipeline with
synthetic_control()
, placebo cases are automatically generated when
constructing the synthetic control of interest. This makes it easy to
explore how unique difference between the observed and synthetic unit is
when compared to the placebos.
smoking_out %>% plot_placebos()
Note that the plot_placebos()
function automatically prunes any
placebos that poorly fit the data in the pre-intervention period. The
reason for doing so is purely visual: those units tend to throw off the
scale when plotting the placebos. To prune, the function looks at the
pre-intervention period mean squared prediction error (MSPE) (i.e. a
metric that reflects how well the synthetic control maps to the observed
outcome time series in pre-intervention period). If a placebo control
has a MSPE that is two times beyond the target case (e.g. “California”),
then it’s dropped. To turn off this behavior, set prune = FALSE
.
smoking_out %>% plot_placebos(prune = FALSE)
Finally, Adabie et al. 2010 outline a way of constructing Fisher’s Exact P-values by dividing the post-intervention MSPE by the pre-intervention MSPE and then ranking all the cases by this ratio in descending order. A p-value is then constructed by taking the rank/total.[^1] The idea is that if the synthetic control fits the observed time series well (low MSPE in the pre-period) and diverges in the post-period (high MSPE in the post-period) then there is a meaningful effect due to the intervention. If the intervention had no effect, then the post-period and pre-period should continue to map onto one another fairly well, yielding a ratio close to 1. If the placebo units fit the data similarly, then we can’t reject the hull hypothesis that there is no effect brought about by the intervention.
This ratio can be easily plotted using plot_mspe_ratio()
, offering
insight into the rarity of the case where the intervention actually
occurred.
smoking_out %>% plot_mspe_ratio()
For more specific information, there is a significance table that can be
extracted with one of the many grab_
prefix functions.
smoking_out %>% grab_significance()
## # A tibble: 39 × 8
## unit_name type pre_mspe post_mspe mspe_ratio rank fishers_exact_pvalue
## <chr> <chr> <dbl> <dbl> <dbl> <int> <dbl>
## 1 California Trea… 3.17 392. 124. 1 0.0256
## 2 Georgia Donor 3.79 179. 47.2 2 0.0513
## 3 Indiana Donor 25.2 770. 30.6 3 0.0769
## 4 West Virginia Donor 9.52 284. 29.8 4 0.103
## 5 Wisconsin Donor 11.1 268. 24.1 5 0.128
## 6 Missouri Donor 3.03 67.8 22.4 6 0.154
## 7 Texas Donor 14.4 277. 19.3 7 0.179
## 8 South Carolina Donor 12.6 234. 18.6 8 0.205
## 9 Virginia Donor 9.81 96.4 9.83 9 0.231
## 10 Nebraska Donor 6.30 52.9 8.40 10 0.256
## # ℹ 29 more rows
## # ℹ 1 more variable: z_score <dbl>
In addition to the main data pipeline for generating the synthetic
control and the plot_
prefix functions for visualizing the output,
there are a number of grab_
prefix functions that offer easy access to
the data contained within a synth pipeline object.
At its core, a synth pipeline is simply a nested tibble data frame, where each component of the synthetic control pipeline is accessible.
smoking_out
## # A tibble: 78 × 11
## .id .placebo .type .outcome .predictors .synthetic_control .unit_weights
## <chr> <dbl> <chr> <list> <list> <list> <list>
## 1 Califor… 0 trea… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 2 Califor… 0 cont… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 3 Alabama 1 trea… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 4 Alabama 1 cont… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 5 Arkansas 1 trea… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 6 Arkansas 1 cont… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 7 Colorado 1 trea… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 8 Colorado 1 cont… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 9 Connect… 1 trea… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## 10 Connect… 1 cont… <tibble> <tibble> <tibble [31 × 3]> <tibble>
## # ℹ 68 more rows
## # ℹ 4 more variables: .predictor_weights <list>, .original_data <list>,
## # .meta <list>, .loss <list>
To access the relevant data fields, the grab_
prefix functions come
into play.
Function | Description |
---|---|
grab_outcome() |
Extract the outcome variable generated by synthetic_control() . |
grab_predictors() |
Extract the aggregate-level covariates generated by generate_predictor() . |
grab_unit_weights() |
Extract the unit weights generated by generate_weights() . |
grab_predictor_weights() |
Extract the predictor variable weights generated by generate_weights() . |
grab_loss() |
Extract the RMSE loss of the optimized weights generated by generate_weights() . |
grab_synthetic_control() |
Extract the synthetic control generated using generate_control() . |
grab_significance() |
Generate inferential statistics comparing the rarity of the unit that actually received the intervention to the placebo units in the donor pool. |
grab_balance_table() |
Compare the distributions of the aggregate-level predictors for the observed intervention unit, the synthetic control, and the donor pool average. |
smoking_out %>% grab_synthetic_control()
## # A tibble: 31 × 3
## time_unit real_y synth_y
## <dbl> <dbl> <dbl>
## 1 1970 123 117.
## 2 1971 121 119.
## 3 1972 124. 124.
## 4 1973 124. 125.
## 5 1974 127. 127.
## 6 1975 127. 127.
## 7 1976 128 128.
## 8 1977 126. 126.
## 9 1978 126. 125.
## 10 1979 122. 123.
## # ℹ 21 more rows
Note that most all the grab_
functions allow for extraction of the
placebo units as well.
smoking_out %>% grab_synthetic_control(placebo = T)
## # A tibble: 1,209 × 5
## .id .placebo time_unit real_y synth_y
## <chr> <dbl> <dbl> <dbl> <dbl>
## 1 California 0 1970 123 117.
## 2 California 0 1971 121 119.
## 3 California 0 1972 124. 124.
## 4 California 0 1973 124. 125.
## 5 California 0 1974 127. 127.
## 6 California 0 1975 127. 127.
## 7 California 0 1976 128 128.
## 8 California 0 1977 126. 126.
## 9 California 0 1978 126. 125.
## 10 California 0 1979 122. 123.
## # ℹ 1,199 more rows
unnest()
…In the current implementation, you can unpack an entire synth pipeline
using unnest()
. The grab_
function is meant to streamline any
specific extraction needs. The entire method is built on top of a
tidyverse infrastructure, so one can side-step most of the package’s
functionality and interact with the synth pipeline output as one would
any nested tbl_df
object.
smoking_out %>%
tidyr::unnest(cols = c(.outcome))
## # A tibble: 1,482 × 50
## .id .placebo .type time_unit California Alabama Arkansas Colorado
## <chr> <dbl> <chr> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 California 0 treated 1970 123 NA NA NA
## 2 California 0 treated 1971 121 NA NA NA
## 3 California 0 treated 1972 124. NA NA NA
## 4 California 0 treated 1973 124. NA NA NA
## 5 California 0 treated 1974 127. NA NA NA
## 6 California 0 treated 1975 127. NA NA NA
## 7 California 0 treated 1976 128 NA NA NA
## 8 California 0 treated 1977 126. NA NA NA
## 9 California 0 treated 1978 126. NA NA NA
## 10 California 0 treated 1979 122. NA NA NA
## # ℹ 1,472 more rows
## # ℹ 42 more variables: Connecticut <dbl>, Delaware <dbl>, Georgia <dbl>,
## # Idaho <dbl>, Illinois <dbl>, Indiana <dbl>, Iowa <dbl>, Kansas <dbl>,
## # Kentucky <dbl>, Louisiana <dbl>, Maine <dbl>, Minnesota <dbl>,
## # Mississippi <dbl>, Missouri <dbl>, Montana <dbl>, Nebraska <dbl>,
## # Nevada <dbl>, `New Hampshire` <dbl>, `New Mexico` <dbl>,
## # `North Carolina` <dbl>, `North Dakota` <dbl>, Ohio <dbl>, Oklahoma <dbl>, …
Spot an issue? Please let me know by posting an issue.
[^1]: Note this implies that you’d need at least 20 cases in the donor pool to get a conventional p-value (.05).