NAP.RdCalculates the non-overlap of all pairs index (Parker & Vannest, 2009).
NAP(
A_data,
B_data,
condition,
outcome,
baseline_phase = NULL,
intervention_phase = NULL,
improvement = "increase",
SE = "unbiased",
confidence = 0.95,
trunc_const = FALSE
)vector of numeric data for A phase. Missing values are dropped.
vector of numeric data for B phase. Missing values are dropped.
vector identifying the treatment condition for each observation in the series.
vector of outcome data for the entire series.
character string specifying which value of
condition corresponds to the baseline phase. Defaults to first
observed value of condition.
character string specifying which value of
condition corresponds to the intervention phase. Defaults to second
unique value of condition.
character string indicating direction of improvement. Default is "increase".
character value indicating which formula to use for calculating the
standard error of NAP, with possible values "unbiased" for the
exactly unbiased estimator, "Hanley" for the Hanley-McNeil
estimator, "null" for the (known) variance under the null hypothesis
of no effect, or "none" to not calculate a standard error. Defaults
to "unbiased".
confidence level for the reported interval estimate. Set to
NULL to omit confidence interval calculations.
logical value indicating whether to return the truncation constant used to calculate the standard error.
A data.frame containing the estimate, standard error, and/or confidence interval.
NAP is calculated as the proportion of all pairs of one observation from each phase in which the measurement from the B phase improves upon the measurement from the A phase, with pairs of data points that are exactly tied being given a weight of 0.5. The range of NAP is [0,1], with a null value of 0.5.
The unbiased variance estimator was described by Sen (1967) and Mee (1990). The Hanley estimator was proposed by Hanley and McNeil (1982). The null variance is a known function of sample size, equal to the exact sampling variance when the null hypothesis of no effect holds. When the null hypothesis does not hold, the null variance will tend to over-estimate the true sampling variance of NAP.
The confidence interval for NAP is calculated based on the symmetrized score-inversion method (Method 5) proposed by Newcombe (2006).
Hanley, J. A., & McNeil, B. J. (1982). The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology, 143, 29–36. doi:doi:10.1148/radiology.143.1.7063747
Mee, W. (1990). Confidence intervals for probabilities and tolerance regions based on a generalization of the Mann-Whitney statistic. Journal of the American Statistical Association, 85(411), 793-800. doi:doi:10.1080/01621459.1990.10474942
Newcombe, R. G. (2006). Confidence intervals for an effect size measure based on the Mann-Whitney statistic. Part 2: Asymptotic methods and evaluation. Statistics in Medicine, 25(4), 559–573. doi:doi:10.1002/sim.2324
Parker, R. I., & Vannest, K. J. (2009). An improved effect size for single-case research: Nonoverlap of all pairs. Behavior Therapy, 40(4), 357–67. doi:doi:10.1016/j.beth.2008.10.006
Sen, P. K. (1967). A note on asymptotically distribution-free confidence bounds for P{X<Y}, based on two independent samples. The Annals of Mathematical Statistics, 29(1), 95-102. https://www.jstor.org/stable/25049448
A <- c(20, 20, 26, 25, 22, 23)
B <- c(28, 25, 24, 27, 30, 30, 29)
NAP(A_data = A, B_data = B)
#> ES Est SE CI_lower CI_upper
#> 1 NAP 0.9166667 0.06900656 0.5973406 0.9860176
# Example from Parker & Vannest (2009)
yA <- c(4, 3, 4, 3, 4, 7, 5, 2, 3, 2)
yB <- c(5, 9, 7, 9, 7, 5, 9, 11, 11, 10, 9)
NAP(yA, yB)
#> ES Est SE CI_lower CI_upper
#> 1 NAP 0.9636364 0.03192622 0.7499741 0.9950729