higgins {heterometa}

Compute heterogeneity statistics after Higgins

Description

Computes various statistics suggested by Higgins and Thompson for quantifying heterogeneity in meta-analysis

Usage

higgins(Q = NULL, k = NULL, I2 = NULL,
   pval = NULL, slab = NULL, conflevel = 95)
## S3 method for class 'higgins'
print(x, type = "I2", na.print = "", ...)

Arguments

Q

Numeric: a vector of heterogeneity \(\chi^2\) from the meta–analyses

k

Numeric: a vector of number of studies in each meta-analysis

I2

Numeric: a vector of values of \(I^2\)

pval

Numeric: a vector of \(p\) values

slab

Character: a vector of labels for the meta-analyses

conflevel

Numeric: a vector of confidence levels

x

An object of class higgins

type

One of "H", "I2", "both"

na.print

What to print instead of NA

...

Argument(s) to be passed through

Details

Either Q or pval or I2 should be provided. If I2 is provided it may be either as a percentage or a proportion. If a single value is given for conflevel it is used for all the analyses.

Limited error checks for illegal parameters are performed. If conflevel is \(<=1\) the function proceeds assuming that was meant as a value of \(\alpha\). If the parameters are supplied as vectors a check is made for equal length. If they are not then a warning is issued but the function tries to return a sensible result which should be checked to see if it is what was desired.

For reference the formulae used are \(Q = H^2 (k - 1)\), \(H^2 = \frac{Q}{k - 1}\), \(I^2 = \frac{H^2 - 1}{H^2}\), \(H^2 = - \frac{1}{I^2 - 1}\).

The print method allows for printing \(H\) or \(I^2\) or both of these and this is controlled by the parameter type.

Value

A list of type higgins containing

H

A data frame with columns Q, k, H, ll, ul, where ll and ul are the confidence limits

I2

A data frame with columns Q, k, I2, ll, ul

call

The call

Note

\(I^2\) is always printed as a percentage even if the input parameter was supplied as a proportion.

Author(s)

Michael Dewey

References

Higgins JPT, Thompson SG (2002). “Quantifying heterogeneity in a meta–analysis.” Statistics in Medicine, 21, 1539–1558. doi:10.1002/sim.1186.

Examples

# first the examples one by one
higgins(14.4, 24) # 1    (1, 1.34)      0 (0, 45)
higgins(14.1, 11) # 1.19 (1, 1.64)     20 (0, 65) probably a typo for 29
higgins(81.5, 19) # 2.13 (1.71, 2.64)  78 (66, 86)
higgins(41.5, 7)  # 2.63 (1.90, 3.65)  86 (72, 92)
higgins(130.3, 3) # 8.07 (6.08, 10.72) 98 (97, 99)
# now repeat getting from dat.higgins02
data(dat.higgins02)
dat <- dat.higgins02
with(dat, higgins(Q, trials, slab = rownames(dat.higgins02)))
# supply I2 or pval
higgins(I2 = dat$I2[3], k = dat$trials[3])
higgins(pval = dat$pval[3], k = dat$trials[3])