Refutations

Refutation checks probe the statistical assumptions underlying a fitted estimator. Call result.refute(data) on any result object that supports it to get a report back.

OLSRefutationReport

class formative.OLSRefutationReport(checks, treatment, outcome)

Results of refutation checks run against an OLS estimation.

Obtain via OLSResult.refute(data).

Example:

result = OLSObservational(
    dag, treatment="education", outcome="income"
).fit(df)
report = result.refute(df)
print(report.summary())

Parameters:

checks (list[RefutationCheck])
treatment (str)
outcome (str)

IVRefutationReport

class formative.IVRefutationReport(checks, treatment, outcome, instrument)

Results of refutation checks run against an IV (2SLS) estimation.

Obtain via IVResult.refute(data). Each check is a RefutationCheck in .checks. The overall verdict is .passed.

Example:

result = IV2SLS(
    dag, treatment="education", outcome="income", instrument="proximity"
).fit(df)
report = result.refute(df)
print(report.summary())

Parameters:

checks (list[RefutationCheck])
treatment (str)
outcome (str)
instrument (str)

MatchingRefutationReport

class formative.MatchingRefutationReport(checks, treatment, outcome)

Results of refutation checks run against a propensity score matching estimation.

Obtain via MatchingResult.refute(data). Each check is a RefutationCheck in .checks. The overall verdict is .passed.

Example:

result = PropensityScoreMatching(
    dag, treatment="education", outcome="income"
).fit(df)
report = result.refute(df)
print(report.summary())

Parameters:

checks (list[RefutationCheck])
treatment (str)
outcome (str)

RCTRefutationReport

class formative.RCTRefutationReport(checks, treatment, outcome)

Results of refutation checks run against an RCT estimation.

Obtain via RCTResult.refute(data).

Example:

result = RCT(dag, treatment="treatment", outcome="outcome").fit(df)
report = result.refute(df)
print(report.summary())

Parameters:

checks (list[RefutationCheck])
treatment (str)
outcome (str)

DiDRefutationReport

class formative.DiDRefutationReport(checks, group, time, outcome)

Results of refutation checks run against a DiD estimation.

Obtain via DiDResult.refute(data).

Example:

result = DiD(dag, group="group", time="time", outcome="outcome").fit(df)
report = result.refute(df)
print(report.summary())

Parameters:

checks (list[RefutationCheck])
group (str)
time (str)
outcome (str)

RDDRefutationReport

class formative.RDDRefutationReport(checks, treatment, running_var, cutoff, outcome)

Results of refutation checks run against an RDD estimation.

Obtain via RDDResult.refute(data).

Example:

result = RDD(dag, treatment="treatment", running_var="score",
             cutoff=0.0, outcome="outcome").fit(df)
report = result.refute(df)
print(report.summary())

Parameters:

checks (list[RefutationCheck])
treatment (str)
running_var (str)
cutoff (float)
outcome (str)

RefutationCheck

class formative.RefutationCheck(name, passed, detail)

Result of a single refutation check.

Parameters:

name (str)
passed (bool)
detail (str)

Assumption

class formative.Assumption(name, testable)

A single modelling assumption required for causal identification.

Every estimator exposes its assumptions via result.assumptions, a list of Assumption objects. Each assumption has a human-readable name and a testable flag indicating whether it can be empirically checked in the data or must be justified on substantive grounds.

Parameters:

name (str)
testable (bool)

name: str: Human-readable description of the assumption.

testable: bool: True if the assumption can be empirically checked; False if it rests on domain knowledge.

fmt_tag()

Return a fixed-width bracketed testability label for use in summary output.

Return type:: str