DALY (version 1.3.0)

sensitivity: DALY sensitivity analysis

Description

This function performs a probabilistic global sensitivity analysis of the overall DALY estimate, based on standardized regression coefficients (method src) or partial correlation coefficients (method pcc). Actual or ranked values may be used, and for method src, coefficients or mapped values may be reported. See below for more details.

Usage

sensitivity(x, method = c("src", "pcc"), rank = FALSE, mapped = TRUE)

## S3 method for class 'DALY_sensitivity':
print(x, digits = 3, signif_stars = getOption("show.signif.stars"), ...)

## S3 method for class 'DALY_sensitivity':
plot(x, alpha = 0.05, main = "Sensitivity analysis",
     show_values = FALSE, value_digits = 3, value_cex = 0.6, \dots)

Arguments

x

Object of class 'DALY', typically obtained from a call to getDALY

method

Sensitivity analysis method: "src" or "pcc". See below for details

rank

Should ranked values be used? Defaults to FALSE

mapped

Should mapped regression coefficients be presented? Defaults to TRUE

digits

Number of decimals to be printed; defaults to 0

signif_stars

Should significance stars be printed?

alpha

Significance level for selecting significant variables; defaults to 0.05

main

Main title of tornado plot; defaults to "Sensitivity analysis"

show_values

Should values be plotted next to bars?; defaults to FALSE

value_digits

Number of significant digits for plotted values; defaults to 3

value_cex

Size of plotted values; defaults to 0.6

...

Additional arguments to be passed to methods

Value

• An object of S3 class DALY_sensitivity, containing the following two elements:
• method
  List containing the specified values for method, rank and mapped
• out
  Either the output of summary.lm (for method src), or a matrix with columns for the partial correlation coefficients (rho) and corresponding p-values (for method pcc).

Details

Sensitivity analysis studies how the uncertainty in the overall DALY estimate can be apportioned to the different sources of uncertainty in the input parameters. These results can therefore help to identify those input parameters that cause significant uncertainty in the overall DALY estimate and that therefore may be the focus of further research if one wishes to reduce the uncertainty in the overall estimate. The sensitivity function implements a probabilistic global sensitivity analysis, in which the analysis is conducted over the full range of plausible input values (hence global), determined by the specified uncertainty distributions (hence probabilistic). Specifying method = "src" will perform a linear regression-based sensitivity analysis. Here, the simulated overall DALY estimates will be regressed against the simulated values for the stochastic input parameters (using lm). To facilitate comparison, the independent terms are standardized such that they are normally distributed with mean zero and standard deviation one (using scale). The resulting regression coefficients are therefore referred to as standardized regression coefficients. Argument rank specifies whether the regression should be performed on the actual values (rank = FALSE; default) or on the ranked values (rank = TRUE). Rank-based regression may be preferred when the relation between output and inputs is non-linear. R^2 values smaller than 0.60 may be indicative of a poor fit of the default linear regression model. If mapped = TRUE, the dependent term is not standardized, such that the resulting mapped regression coefficients correspond to the change in overall DALY given one standard deviation change in the corresponding input parameter. If mapped = FALSE, the dependent term is standardized, such that the resulting standardized regression coefficients correspond to the number of standard deviations change in overall DALY given one standard deviation change in the corresponding input parameter. Specifying method = "pcc" will calculate partial correlation coefficients for each of the input variables. Partial correlation coefficients represent the correlation between two variables when adjusting for other variables. In the presence of important interactions between input variables, partial correlation coefficients may be preferred over standardized regression coefficients. Argument rank specifies whether the correlation should be calculated between the actual values (rank = FALSE; default) or between the ranked values (rank = TRUE). Method plot can be used to generate a tornado plot of the significant input variables.

References

• Vose D (2000).Risk analysis: a quantitative guide. John Wiley & Sons.

Saltelli A, Chan K, Scott EM (2000). Sensitivity analysis. John Wiley & Sons.

See Also

Package sensitivity, for more advanced sensitivity analysis methods. getDALY lm cor.test

Examples

##= load NCC example ======================================
setDALYexample(1)

##= perform DALY calculation, store results in 'x' ========
x <- getDALY()

##= perform DALY sensitivity analysis =====================

## default sensitivity analysis
sa <- sensitivity(x)
print(sa)

## tornado plot of default sensitivity analysis
plot(sa, show_values = TRUE)

## standardized regression coefficients -------------------
sensitivity(x, method = "src", rank = FALSE, mapped = TRUE)
sensitivity(x, method = "src", rank = FALSE, mapped = FALSE)

## standardized rank regression coefficients --------------
sensitivity(x, method = "src", rank = TRUE, mapped = TRUE)
sensitivity(x, method = "src", rank = TRUE, mapped = FALSE)

## partial correlation coefficients -----------------------
sensitivity(x, method = "pcc", rank = FALSE)

## partial rank correlation coefficients ------------------
sensitivity(x, method = "pcc", rank = FALSE)