Package 'ptetools'

The estimated overall ATT

Standard error for overall ATT

The type of aggregation to be done. Default is "overall".

Holds the length of exposure (for dynamic effects), the group (for selective treatment timing), or the time period (for calendar time effects)

The ATT specific to egt

The standard error specific to egt

A critical value for computing uniform confidence bands for dynamic effects, selective treatment timing, or time period effects.

The influence function of the chosen aggregated parameters

The minimum event time computed in the event study results. This is useful when there are a huge number of pre-treatment periods.

The maximum event time computed in the event study results. This is useful when there are a huge number of post-treatment periods.

Drops groups that do not have at least balance_e periods of post-treatment data. This keeps the composition of groups constant across different event times in an event study. Default is NULL, in which case this is ignored.

A DIDparams object

group-time average treatment effect

influence function

A place to return anything extra from particular group-time average treatment effect calculations. For DID, this might be something like propensity score estimates, regressions of untreated potential outcomes on covariates. For ife, this could be something like the first step regression 2sls estimates. This argument is also potentially useful for debugging.

group-time average treatment effect

A place to return anything extra from particular group-time average treatment effect calculations. For DID, this might be something like propensity score estimates, regressions of untreated potential outcomes on covariates. For ife, this could be something like the first step regression 2sls estimates. This argument is also potentially useful for debugging.

list of attgt results from compute.pte

pte_params object

pte_params object

This is a function that should take in data, g (for group), tp (for time period), and ... and be able to return the appropriate data.frame that can be used by attgt_fun to produce ATT(g=g,t=tp). The data frame should be constructed using gt_data_frame in order to guarantee that it has the appropriate columns that identify which group an observation belongs to, etc.

This is a function that should work in the case where there is a single group and the "right" number of time periods to recover an estimate of the ATT. For example, in the contest of difference in differences, it would need to work for a single group, find the appropriate comparison group (untreated units), find the right time periods (pre- and post-treatment), and then recover an estimate of ATT for that group. It will be called over and over separately by groups and by time periods to compute ATT(g,t)'s.

The function needs to work in a very specific way. It should take in the arguments: data, .... data should be constructed using the function gt_data_frame which checks to make sure that data has the correct columns defined. ... are additional arguments (such as formulas for covariates) that attgt_fun needs. From these arguments attgt_fun must return a list with element ATT containing the group-time average treatment effect for that group and that time period.

If attgt_fun returns an influence function (which should be provided in a list element named inf_func), then the code will use the multiplier bootstrap to compute standard errors for group-time average treatment effects, an overall treatment effect parameter, and a dynamic treatment effect parameter (i.e., event study parameter). If attgt_fun does not return an influence function, then the same objects will be computed using the empirical bootstrap. This is usually (perhaps substantially) easier to code, but also will usually be (perhaps substantially) computationally slower.

extra arguments that can be passed to create the correct subsets of the data (depending on subset_fun), to estimate group time average treatment effects (depending on attgt_fun), or to aggregating treatment effects (particularly useful are min_e, max_e, and balance_e arguments to event study aggregations)

the critical value

the significance level

data that is "local" to a particular group-time average treatment effect

one-sided formula for covariates used in the propensity score and outcome regression models

extra function arguments; not used here

vector containing the values of the dose used in estimation

estimate of the overall ATT, the mean of ATT(D) given D > 0

the standard error of the estimate of overall_att

the influence function for estimating overall_att

estimate of the overall ACRT, the mean of ACRT(D|D) given D > 0

the standard error for the estimate of overall_acrt

the influence function for estimating overall_acrt

estimates of ATT(d) for each value of dose

standard error of ATT(d) for each value of dose

critical value to produce pointwise or uniform confidence interval for ATT(d)

matrix containing the influence function from estimating ATT(d)

estimates of ACRT(d) for each value of dose

standard error of ACRT(d) for each value of dose

critical value to produce pointwise or uniform confidence interval for ACRT(d)

matrix containing the influence function from estimating ACRT(d)

a pte_params object containing other parameters passed to the function

A pte_results object

a dose_obj that holds results with a continuous treatment

whether to plot ATT(d) or ACRT(d), defaults to att for plotting ATT(d). For ACRT(d), use "acrt"

numeric vector of groups for ATT(g,t)

numeric vector of time periods for ATT(g,t)

numeric vector containing the value of ATT(g,t) for corresponding group and time period

analytical asymptotic variance matrix for ATT(g,t)'s

numeric vector of standard errors

critical value (usually a critical value for conducting uniform inference)

matrix of influence function

number of unique individuals

Wald statistic for ATT(g,t) version of pre-test of parallel trends assumption

p-value for Wald pre-test of ATT(g,t) version of parallel trends assumption

logical indicating whether or not to report a confidence band

significance level

pte_params object

list containing extra returns at the group-time level

data that will be checked to see if has right format for computing group-time average treatment effects

the full dataset

the current group

the current time period

additional arguments

the full dataset

the current group

the current time period

extra arguments to get the subset correct

influence function matrix

number of bootstrap iterations; default is 100

significance level; default is 0.05

A group_time_att object to be aggregated

Drops groups that do not have at least balance_e periods of post-treatment data. This keeps the composition of groups constant across different event times in an event study. Default is NULL, in which case this is ignored.

The minimum event time computed in the event study results. This is useful when there are a huge number of pre-treatment periods.

The maximum event time computed in the event study results. This is useful when there are a huge number of post-treatment periods.

extra arguments

list of attgt results from compute.pte

pte_params object

This is a function that should take in data, yname (the name of the outcome variable in data), gname (the name of the group variable), idname (the name of the id variable), and possibly other arguments such as the significance level alp, the number of bootstrap iterations biters, and how many clusters for parallel computing in the bootstrap cl. The key thing that needs to be figured out in this function is which groups and time periods ATT(g,t) should be computed in. The function should return a pte_params object which contains all of the parameters passed into the function as well as glist and tlist which should be ordered lists of groups and time periods for ATT(g,t) to be computed.

This function provides also provides a good place for error handling related to the types of data that can be handled.

The pte package contains the function setup_pte that is a lightweight function that basically just takes the data, omits the never-treated group from glist but includes all other groups and drops the first time period. This works in cases where ATT would be identified in the 2x2 case (i.e., where there are two time periods, no units are treated in the first period and the identification strategy "works" with access to a treated and untreated group and untreated potential outcomes for both groups in the first period) — for example, this approach works if DID is the identification strategy.

This is a function that should take in data, g (for group), tp (for time period), and ... and be able to return the appropriate data.frame that can be used by attgt_fun to produce ATT(g=g,t=tp). The data frame should be constructed using gt_data_frame in order to guarantee that it has the appropriate columns that identify which group an observation belongs to, etc.

This is a function that should work in the case where there is a single group and the "right" number of time periods to recover an estimate of the ATT. For example, in the contest of difference in differences, it would need to work for a single group, find the appropriate comparison group (untreated units), find the right time periods (pre- and post-treatment), and then recover an estimate of ATT for that group. It will be called over and over separately by groups and by time periods to compute ATT(g,t)'s.

The function needs to work in a very specific way. It should take in the arguments: data, .... data should be constructed using the function gt_data_frame which checks to make sure that data has the correct columns defined. ... are additional arguments (such as formulas for covariates) that attgt_fun needs. From these arguments attgt_fun must return a list with element ATT containing the group-time average treatment effect for that group and that time period.

If attgt_fun returns an influence function (which should be provided in a list element named inf_func), then the code will use the multiplier bootstrap to compute standard errors for group-time average treatment effects, an overall treatment effect parameter, and a dynamic treatment effect parameter (i.e., event study parameter). If attgt_fun does not return an influence function, then the same objects will be computed using the empirical bootstrap. This is usually (perhaps substantially) easier to code, but also will usually be (perhaps substantially) computationally slower.

A place to return anything extra from particular group-time average treatment effect calculations. For DID, this might be something like propensity score estimates, regressions of untreated potential outcomes on covariates. For ife, this could be something like the first step regression 2sls estimates. This argument is also potentially useful for debugging.

extra arguments that can be passed to create the correct subsets of the data (depending on subset_fun), to estimate group time average treatment effects (depending on attgt_fun), or to aggregating treatment effects (particularly useful are min_e, max_e, and balance_e arguments to event study aggregations)

ATT(g,t)'s

pte_params object

list of group-time specific results

pte_params object

extra arguments

Name of outcome in data

Name of group in data

Name of time period in data

Name of id in data

balanced panel data

This is a function that should take in data, yname (the name of the outcome variable in data), gname (the name of the group variable), idname (the name of the id variable), and possibly other arguments such as the significance level alp, the number of bootstrap iterations biters, and how many clusters for parallel computing in the bootstrap cl. The key thing that needs to be figured out in this function is which groups and time periods ATT(g,t) should be computed in. The function should return a pte_params object which contains all of the parameters passed into the function as well as glist and tlist which should be ordered lists of groups and time periods for ATT(g,t) to be computed.

This function provides also provides a good place for error handling related to the types of data that can be handled.

The pte package contains the function setup_pte that is a lightweight function that basically just takes the data, omits the never-treated group from glist but includes all other groups and drops the first time period. This works in cases where ATT would be identified in the 2x2 case (i.e., where there are two time periods, no units are treated in the first period and the identification strategy "works" with access to a treated and untreated group and untreated potential outcomes for both groups in the first period) — for example, this approach works if DID is the identification strategy.

This is a function that should take in data, g (for group), tp (for time period), and ... and be able to return the appropriate data.frame that can be used by attgt_fun to produce ATT(g=g,t=tp). The data frame should be constructed using gt_data_frame in order to guarantee that it has the appropriate columns that identify which group an observation belongs to, etc.

This is a function that should work in the case where there is a single group and the "right" number of time periods to recover an estimate of the ATT. For example, in the contest of difference in differences, it would need to work for a single group, find the appropriate comparison group (untreated units), find the right time periods (pre- and post-treatment), and then recover an estimate of ATT for that group. It will be called over and over separately by groups and by time periods to compute ATT(g,t)'s.

The function needs to work in a very specific way. It should take in the arguments: data, .... data should be constructed using the function gt_data_frame which checks to make sure that data has the correct columns defined. ... are additional arguments (such as formulas for covariates) that attgt_fun needs. From these arguments attgt_fun must return a list with element ATT containing the group-time average treatment effect for that group and that time period.

If attgt_fun returns an influence function (which should be provided in a list element named inf_func), then the code will use the multiplier bootstrap to compute standard errors for group-time average treatment effects, an overall treatment effect parameter, and a dynamic treatment effect parameter (i.e., event study parameter). If attgt_fun does not return an influence function, then the same objects will be computed using the empirical bootstrap. This is usually (perhaps substantially) easier to code, but also will usually be (perhaps substantially) computationally slower.

whether or not to report a uniform (instead of pointwise) confidence band (default is TRUE)

significance level; default is 0.05

should be one of "multiplier" (the default) or "empirical". The multiplier bootstrap is generally much faster, but attgt_fun needs to provide an expression for the influence function (which could be challenging to figure out). If no influence function is provided, then the pte package will use the empirical bootstrap no matter what the value of this parameter.

The name of the column that contains sampling weights. The default is NULL, in which case no sampling weights are used.

which type of group-time effects are computed. The default is "att". Different estimation strategies can implement their own choices for gt_type

For functions that compute quantile treatment effects, this is a specific quantile at which to report results, e.g., ret_quantile = 0.5 will return that the qte at the median.

Logical indicating whether or not untreated potential outcomes can be estimated in one shot, i.e., for all groups and time periods. Main use case would be for one-shot imputation estimators. Not supported yet.

Logical indicating whether or not untreated potential outcomes can be estimated for all groups in the same time period. Not supported yet.

Logical indicating whether or not untreated potential outcomes can be estimated for all time periods for a single group. Not supported yet. These functions aim at reducing or eliminating running the same code multiple times.

An optional function to customize results when the gt-specific function returns the distribution of treated and untreated potential outcomes. The default is process_dtt_gt, which is a function provided by the package. See that function for an example of what this function should return. This is unused is unused except in cases where the results involve distributions.

An optional function to customize results when the gt-specific function returns treatment effects that depend on dose (i.e., amount of the treatment). The default is process_dose_gt, which is a function provided by the package. See that function for an example of what this function should return. This is unused except in cases where the results involve doses.

number of bootstrap iterations; default is 100

number of clusters to be used when bootstrapping; default is 1

keeps track of through the call from external functions/packages

extra arguments that can be passed to create the correct subsets of the data (depending on subset_fun), to estimate group time average treatment effects (depending on attgt_fun), or to aggregating treatment effects (particularly useful are min_e, max_e, and balance_e arguments to event study aggregations)

A group_time_att object to be aggregated

The type of aggregation to be done. Default is "overall".

Drops groups that do not have at least balance_e periods of post-treatment data. This keeps the composition of groups constant across different event times in an event study. Default is NULL, in which case this is ignored.

The minimum event time computed in the event study results. This is useful when there are a huge number of pre-treatment periods.

The maximum event time computed in the event study results. This is useful when there are a huge number of post-treatment periods.

extra arguments

data that is "local" to a particular group-time average treatment effect

one-sided formula for covariates used in the propensity score and outcome regression models

Whether or not to take the first difference of the outcome. The default is FALSE. To use difference-in-differences, set this to be TRUE.

A formula for time varying covariates to enter the first estimation step models. The default is not to include any, and, hence, to only include pre-treatment covariates.

Whether to include the lagged outcome as a covariate. Default is FALSE.

Which type of estimation method to use. Default is "dr" for doubly robust. The other option is "reg" for regression adjustment.

extra function arguments; not used here

Name of outcome in data

Name of group in data

Name of time period in data

Name of id in data

balanced panel data

one-sided formula for covariates used in the propensity score and outcome regression models

Whether or not to take the first difference of the outcome. The default is FALSE. To use difference-in-differences, set this to be TRUE.

A formula for time varying covariates to enter the first estimation step models. The default is not to include any, and, hence, to only include pre-treatment covariates.

Whether to include the lagged outcome as a covariate. Default is FALSE.

Which type of estimation method to use. Default is "dr" for doubly robust. The other option is "reg" for regression adjustment.

how many periods before the treatment actually takes place that it can have an effect on outcomes

The type of base period to use. This only affects the numeric value of results in pre-treatment periods. Results in post-treatment periods are not affected by this choice. The default is "varying", where the base period will "back up" to the immediately preceding period in pre-treatment periods. The other option is "universal" where the base period is fixed in pre-treatment periods to be the period right before the treatment starts. "Universal" is commonly used in difference-in-differences applications, but can be unnatural for other identification strategies.

Which group is used as the comparison group. The default choice is "notyettreated", but different estimation strategies can implement their own choices for the control group

The name of the column that contains sampling weights. The default is NULL, in which case no sampling weights are used.

whether or not to report a uniform (instead of pointwise) confidence band (default is TRUE)

significance level; default is 0.05

should be one of "multiplier" (the default) or "empirical". The multiplier bootstrap is generally much faster, but attgt_fun needs to provide an expression for the influence function (which could be challenging to figure out). If no influence function is provided, then the pte package will use the empirical bootstrap no matter what the value of this parameter.

number of bootstrap iterations; default is 100

number of clusters to be used when bootstrapping; default is 1

attgt results

vector of doses

ATT(d) for each value of dose

ACRT(d) for each value of dose

a pte_params object

data.frame holding attgt results

data.frame holding overall results

data.frame holding group results

data.frame holding dynamic results

vector containing weights on underlying ATT(g,t) for overall treatment effect parameter

list containing weights on underlying ATT(g,t) for each value of e corresponding to the dynamic treatment effect parameters.

list containing weights on underlying ATT(g,t) corresponding to deliver averaged group-specific treatment effects

A place to return anything extra from particular group-time average treatment effect calculations. For DID, this might be something like propensity score estimates, regressions of untreated potential outcomes on covariates. For ife, this could be something like the first step regression 2sls estimates. This argument is also potentially useful for debugging.

Name of outcome in data

Name of group in data

Name of time period in data

Name of id in data

balanced panel data

list of groups to create group-time average treatment effects for

list of time periods to create group-time average treatment effects for

whether or not to report a uniform (instead of pointwise) confidence band (default is TRUE)

significance level; default is 0.05

which type of bootstrap to use

how many periods before the treatment actually takes place that it can have an effect on outcomes

The type of base period to use. This only affects the numeric value of results in pre-treatment periods. Results in post-treatment periods are not affected by this choice. The default is "varying", where the base period will "back up" to the immediately preceding period in pre-treatment periods. The other option is "universal" where the base period is fixed in pre-treatment periods to be the period right before the treatment starts. "Universal" is commonly used in difference-in-differences applications, but can be unnatural for other identification strategies.

The name of the column that contains sampling weights. The default is NULL, in which case no sampling weights are used.

Which group is used as the comparison group. The default choice is "notyettreated", but different estimation strategies can implement their own choices for the control group

which type of group-time effects are computed. The default is "att". Different estimation strategies can implement their own choices for gt_type

For functions that compute quantile treatment effects, this is a specific quantile at which to report results, e.g., ret_quantile = 0.5 will return that the qte at the median.

Logical indicating whether or not untreated potential outcomes can be estimated in one shot, i.e., for all groups and time periods. Main use case would be for one-shot imputation estimators. Not supported yet.

Logical indicating whether or not untreated potential outcomes can be estimated for all groups in the same time period. Not supported yet.

Logical indicating whether or not untreated potential outcomes can be estimated for all time periods for a single group. Not supported yet. These functions aim at reducing or eliminating running the same code multiple times.

number of bootstrap iterations; default is 100

number of clusters to be used when bootstrapping; default is 1

keeps track of through the call from external functions/packages

attgt results

overall_att results

event_study results

pte_params object

list of attgt results from compute.pte

pte_params object

A place to return anything extra from particular group-time average treatment effect calculations. For DID, this might be something like propensity score estimates, regressions of untreated potential outcomes on covariates. For ife, this could be something like the first step regression 2sls estimates. This argument is also potentially useful for debugging.

list of attgt results from compute.pte

pte_params object

A place to return anything extra from particular group-time average treatment effect calculations. For DID, this might be something like propensity score estimates, regressions of untreated potential outcomes on covariates. For ife, this could be something like the first step regression 2sls estimates. This argument is also potentially useful for debugging.

Name of outcome in data

Name of group in data

Name of time period in data

Name of id in data

balanced panel data

The number of required pre-treatment periods to implement the estimation strategy. Default is 1.

how many periods before the treatment actually takes place that it can have an effect on outcomes

The type of base period to use. This only affects the numeric value of results in pre-treatment periods. Results in post-treatment periods are not affected by this choice. The default is "varying", where the base period will "back up" to the immediately preceding period in pre-treatment periods. The other option is "universal" where the base period is fixed in pre-treatment periods to be the period right before the treatment starts. "Universal" is commonly used in difference-in-differences applications, but can be unnatural for other identification strategies.

whether or not to report a uniform (instead of pointwise) confidence band (default is TRUE)

significance level; default is 0.05

which type of bootstrap to use

The name of the column that contains sampling weights. The default is NULL, in which case no sampling weights are used.

which type of group-time effects are computed. The default is "att". Different estimation strategies can implement their own choices for gt_type

For functions that compute quantile treatment effects, this is a specific quantile at which to report results, e.g., ret_quantile = 0.5 will return that the qte at the median.

number of bootstrap iterations; default is 100

number of clusters to be used when bootstrapping; default is 1

keeps track of through the call from external functions/packages

additional arguments

Name of outcome in data

Name of group in data

Name of time period in data

Name of id in data

balanced panel data

whether or not to report a uniform (instead of pointwise) confidence band (default is TRUE)

significance level; default is 0.05

which type of bootstrap to use

which type of group-time effects are computed. The default is "att". Different estimation strategies can implement their own choices for gt_type

For functions that compute quantile treatment effects, this is a specific quantile at which to report results, e.g., ret_quantile = 0.5 will return that the qte at the median.

number of bootstrap iterations; default is 100

number of clusters to be used when bootstrapping; default is 1

keeps track of through the call from external functions/packages

additional arguments

the full dataset

the current group

the current time period

whether to use "notyettreated" (default) or "nevertreated"

the number of periods of anticipation (i.e., number of periods before the treatment happens where the treatment can "already" affect the outcome)

The type of base period to use. This only affects the numeric value of results in pre-treatment periods. Results in post-treatment periods are not affected by this choice. The default is "varying", where the base period will "back up" to the immediately preceding period in pre-treatment periods. The other option is "universal" where the base period is fixed in pre-treatment periods to be the period right before the treatment starts. "Universal" is commonly used in difference-in-differences applications, but can be unnatural for other identification strategies.

extra arguments to get the subset correct