Skip to contents

Tests funnel asymmetry directly from an ordinary MARS fit, without fitting a selection model. Retains the parent mean design and clusters inference by study.

Usage

egger_test(
  object,
  robust_args = list(type = "CR1S", inference = "naive-t"),
  cluster = NULL
)

# S3 method for class 'mars_egger'
print(x, digits = 4L, ...)

Arguments

object

Ordinary mars fit, or pub_bias fit whose ordinary parent is used. For network models use network_bias_test(). Penalized fits are unsupported.

robust_args

Named controls for the shared robust engine. Defaults to CR1S covariance and G-1 t inference. See mars() for available alternatives.

cluster

Optional column naming independent clusters. NULL uses the parent's robustID cluster when stored, otherwise studyID. This determines inference groups, not the regression weights. Missing cluster identifiers are rejected.

x

Egger-test result.

digits

Significant digits for printing.

...

Unsupported print arguments are rejected.

Value

A mars_egger list with ok, asymmetry intercept, SE, statistic, reference degrees of freedom, p_value, complete coefficient/design/covariance information, study counts and interpretation diagnostics. Unavailable tests have ok=FALSE and a reason. Print displays the asymmetry test and diagnostic notes.

Details

Fits \(y/\mathrm{SE}\) on a constant asymmetry intercept and the original mean design divided by sampling SE, using ordinary least squares. No additional inverse- variance weights are applied on this standardized scale. With an intercept-only parent this equals raw-effect regression on sampling SE with inverse-variance weights. QR computation scales design columns for numerical stability. The null is a zero asymmetry coefficient, not absence of publication bias. Moderators are retained; study-cluster robust inference accommodates dependence within studies but does not remove confounding by unmodeled effect modifiers. Fewer than ten independent studies and little SE variation are flagged. Ten is a rule of thumb. Rank deficiency, invalid SEs or reference degrees of freedom make inference unavailable. Original Egger-type tests can show artefactual asymmetry for SMDs and odds ratios because effects and SEs can be coupled. Nonsignificance does not exclude missing evidence. Assess effect-measure suitability and interpret alongside study characteristics and funnel plots.

References

Egger et al. (1997). Bias in meta-analysis detected by a simple, graphical test. BMJ, 315, 629-634. doi:10.1136/bmj.315.7109.629 .

Examples

fit <- mars(bcg,studyID="trial",effectID=NULL,sample_size=NULL,formula=yi~1,
  variance="vi",varcov_type="univariate",structure="univariate")
egger_test(fit)
#> Egger-type small-study-effects test
#> Egger-type small-study-effects regression with study-cluster robust SE (CR1S, naive-t) 
#>  asymmetry    SE statistic df p_value
#>     -2.112 1.278    -1.653 12  0.1242
#> Independent clusters: 13 | Effects: 13 
#> Asymmetry does not establish publication bias.