BinomRatioCI: Confidence Intervals for the Ratio of Binomial and Multinomial Proportions

Description

A number of methods have been develeloped for obtaining confidence intervals for the ratio of two binomial proportions. These include the Wald/Katz-log method (Katz et al. 1978), adjusted-log (Walter 1975, Pettigrew et al. 1986), Koopman asymptotic score (Koopman 1984), Inverse hyperbolic sine transformation (Newman 2001), the Bailey method (Bailey (1987), and the Noether (1957) procedure. Koopman results are found iteratively for most intervals using root finding.

Usage

BinomRatioCI(x1, n1, x2, n2, conf.level = 0.95, method = "katz.log", 
             bonf = FALSE, tol = .Machine$double.eps^0.25, R = 1000, r = length(x1))

Arguments

The ratio numerator number of successes. A scalar or vector.

The ratio numerator number of trials. A scalar or vector of length(y1)

The ratio denominator number of successes. A scalar or vector of length(y1)

The ratio denominator number of trials. A scalar or vector of length(y1)

conf.level

The level of confidence, i.e. 1 - P(type I error).

method

Confidence interval method. One of "adj.log", "bailey", "boot", "katz.log", "koopman", "sinh-1" or "noether". Partial distinct names can be used.

bonf

Logical, indicating whether or not Bonferroni corrections should be applied for simultaneous inference if y1, y2, n1 and n2 are vectors.

tol

The desired accuracy (convergence tolerance) for the iterative root finding procedure when finding Koopman intevals. The default is taken to be the smallest positive floating-point number of the workstation implementing the function, raised to the 0.25 power, and will normally be approximately 0.0001.

If method "boot" is chosen, the number of bootstrap iterations.

The number of ratios to which family-wise inferences are being made. Assumed to be length(y1).

Value

Returns a list of class = "ci". Default output is a matrix with the point and interval estimate.

Details

Let $Y_1$ and $Y_2$ be multinomial random variables with parameters $n_1, \pi_{1i}$, and $n_2, \pi_{2i}$, respectively; where $i = \{1, 2, 3, \dots, r\}$. This encompasses the binomial case in which $r = 1$. We define the true selection ratio for the ith resource of r total resources to be: $$\theta_{i}=\frac{\pi _{1i}}{\pi _{2i}}$$

where $\pi_{1i}$ and $\pi_{2i}$ represent the proportional use and availability of the ith resource, respectively. Note that if $r = 1$ the selection ratio becomes relative risk. The maximum likelihood estimators for $\pi_{1i}$ and $\pi_{2i}$ are the sample proportions:

$${{\hat{\pi }}_{1i}}=\frac{{{y}_{1i}}}{{{n}_{1}}},$$ and $${{\hat{\pi }}_{2i}}=\frac{{{y}_{2i}}}{{{n}_{2}}}$$

where $y_{1i}$ and $y_{2i}$ are the observed counts for use and availability for the ith resource. The estimator for $\theta_i$ is:

$$\hat{\theta}_{i}=\frac{\hat{\pi}_{1i}}{\hat{\pi }_{2i}}.$$

Method	Algorithm

Katz-log	$\hat\theta_i\times$ exp$(\pm z_1-\alpha/2\hat{\sigma}_W)$,
	where $\hat\sigma_W^2=\frac{(1-\hat{\pi} _{1i})}{\hat{\pi}_{1i}n_1}+\frac{(1-\hat{\pi}_{2i})}{\hat{\pi}_{2i}n_2}$.

Adjusted-log	$\hat{\theta}_{Ai}\times$ exp$(\pm z_1-\alpha /2\hat{\sigma}_A)$,
	where $\hat{\theta}_{Ai}=\frac{y_{1i}+0.5/n_1+0.5}{y_{2i}+0.5/n_2+0.5}$,
	$\hat{\sigma}_A^2=\frac{1}{y_1+0.5}-\frac{1}{n_1+0.5}+\frac{1}{y_2+0.5}-\frac{1}{n_2+0.5}$.

Bailey	$\hat{\theta} _i\left[\frac{1\pm z_1-\left( \alpha /2 \right)\left( \hat{\pi}_{1i}'/y_{1i}+\hat{\pi}_{2i}'/y_{2i}-z_1-\left(\alpha/2 \right)^2\hat{\pi} _{1i}'\hat{\pi}_{2i}'/9y_{1i}y_{2i} \right)^{1/2}/3}{1-z_{1-\left(\alpha/2 \right)^2}\hat{\pi} _{2i}'/9y_{2i}} \right]^3$,
	where $\hat{\pi_{1i}}'$ = 1 - $\hat{\pi}_{1i}$, and $\hat{\pi}_{2i}'$ = 1 - $\hat{\pi}_{2i}$.

Inv. hyperbolic sine	$\ln({{\hat{\theta }}_{i}})\pm \left[ 2sin{{h}^{-1}}\left( \frac{{{z}_{(1-\alpha /2)}}}{2}\sqrt{\frac{1}{{{y}_{1i}}}-\frac{1}{{{n}_{1}}}+\frac{1}{{{y}_{2i}}}-\frac{1}{{{n}_{2}}}} \right) \right]$,

Koopman	Find $X^2(\theta_0)$ = $\chi _1^2(1 - \alpha)$, where
	${{\tilde{\pi }}_{1i}}=\frac{{{\theta }_{0}}({{n}_{1}}+{{y}_{2i}})+{{y}_{1i}}+{{n}_{2}}-{{[{{\{{{\theta }_{0}}({{n}_{1}}+{{y}_{2i}})+{{y}_{1i}}+ {{n}_{2}}\}}^{2}}-4{{\theta }_{0}}({{n}_{1}}+{{n}_{2}})({{y}_{1i}}+{{y}_{2i}})]}^{0.5}}}{2({{n}_{1}}+{{n}_{2}})}$,
	${{\tilde{\pi }}_{2i}}=\frac{{{{\tilde{\pi }}}_{1i}}}{{{\theta }_{0}}}$, and ${{X}^{2}}({{\theta}_{0}})=\frac{{{\left( {{y}_{1i}}-{{n}_{1}}{{{\tilde{\pi }}}_{1i}} \right)}^{2}}} {{{n}_{1}}{{{\tilde{\pi }}}_{1i}}(1-{{{\tilde{\pi }}}_{1i}})}\left\{ 1+\frac{{{n}_{1}}({{\theta}_{0}}-{{{\tilde{\pi }}}_{1i}})}{{{n}_{2}}(1-{\tilde{\pi}_{1i}})} \right\}$.

Noether	$\hat{\theta}_i\pm z_1-\alpha/2\hat{\sigma}_N$,

Exception handling strategies are generally necessary in the cases $y_1$ = 0, $n_1$ = $y_1$, $y_2$ = 0, and $n_2$ = $y_2$ (see Aho and Bowyer, in review).

The bootstrap method currently employs percentile confidence intervals.

References

Agresti, A., Min, Y. (2001) On small-sample confidence intervals for parameters in discrete distributions. Biometrics 57: 963-97.

Aho, K., and Bowyer, T. (In review) Confidence intervals for ratios of multinomial proportions: implications for selection ratios. Methods in Ecology and Evolution.

Bailey, B.J.R. (1987) Confidence limits to the risk ratio. Biometrics 43(1): 201-205.

Katz, D., Baptista, J., Azen, S. P., and Pike, M. C. (1978) Obtaining confidence intervals for the risk ratio in cohort studies. Biometrics 34: 469-474

Koopman, P. A. R. (1984) Confidence intervals for the ratio of two binomial proportions. Biometrics 40:513-517.

Manly, B. F., McDonald, L. L., Thomas, D. L., McDonald, T. L. and Erickson, W.P. (2002) Resource Selection by Animals: Statistical Design and Analysis for Field Studies. 2nd edn. Kluwer, New York, NY

Newcombe, R. G. (2001) Logit confidence intervals and the inverse sinh transformation. The American Statistician 55: 200-202.

Pettigrew H. M., Gart, J. J., Thomas, D. G. (1986) The bias and higher cumulants of the logarithm of a binomial variate. Biometrika 73(2): 425-435.

Walter, S. D. (1975) The distribution of Levins measure of attributable risk. Biometrika 62(2): 371-374.

Examples

Run this code

# NOT RUN {
# From Koopman (1984)
BinomRatioCI(x1 = 36, n1 = 40, x2 = 16, n2 = 80, method = "katz")
BinomRatioCI(x1 = 36, n1 = 40, x2 = 16, n2 = 80, method = "koop")
# }

Run the code above in your browser using DataLab

Method	Algorithm

Katz-log	\(\hat\theta_i\times\) exp\((\pm z_1-\alpha/2\hat{\sigma}_W)\),
	where \(\hat\sigma_W^2=\frac{(1-\hat{\pi} _{1i})}{\hat{\pi}_{1i}n_1}+\frac{(1-\hat{\pi}_{2i})}{\hat{\pi}_{2i}n_2}\).

Adjusted-log	\(\hat{\theta}_{Ai}\times\) exp\((\pm z_1-\alpha /2\hat{\sigma}_A)\),
	where \(\hat{\theta}_{Ai}=\frac{y_{1i}+0.5/n_1+0.5}{y_{2i}+0.5/n_2+0.5}\),
	\(\hat{\sigma}_A^2=\frac{1}{y_1+0.5}-\frac{1}{n_1+0.5}+\frac{1}{y_2+0.5}-\frac{1}{n_2+0.5}\).

Bailey	\(\hat{\theta} _i\left[\frac{1\pm z_1-\left( \alpha /2 \right)\left( \hat{\pi}_{1i}'/y_{1i}+\hat{\pi}_{2i}'/y_{2i}-z_1-\left(\alpha/2 \right)^2\hat{\pi} _{1i}'\hat{\pi}_{2i}'/9y_{1i}y_{2i} \right)^{1/2}/3}{1-z_{1-\left(\alpha/2 \right)^2}\hat{\pi} _{2i}'/9y_{2i}} \right]^3\),
	where \(\hat{\pi_{1i}}'\) = 1 - \(\hat{\pi}_{1i}\), and \(\hat{\pi}_{2i}'\) = 1 - \(\hat{\pi}_{2i}\).

Inv. hyperbolic sine	\(\ln({{\hat{\theta }}_{i}})\pm \left[ 2sin{{h}^{-1}}\left( \frac{{{z}_{(1-\alpha /2)}}}{2}\sqrt{\frac{1}{{{y}_{1i}}}-\frac{1}{{{n}_{1}}}+\frac{1}{{{y}_{2i}}}-\frac{1}{{{n}_{2}}}} \right) \right]\),

Koopman	Find \(X^2(\theta_0)\) = \(\chi _1^2(1 - \alpha)\), where
	\({{\tilde{\pi }}_{1i}}=\frac{{{\theta }_{0}}({{n}_{1}}+{{y}_{2i}})+{{y}_{1i}}+{{n}_{2}}-{{[{{\{{{\theta }_{0}}({{n}_{1}}+{{y}_{2i}})+{{y}_{1i}}+ {{n}_{2}}\}}^{2}}-4{{\theta }_{0}}({{n}_{1}}+{{n}_{2}})({{y}_{1i}}+{{y}_{2i}})]}^{0.5}}}{2({{n}_{1}}+{{n}_{2}})}\),
	\({{\tilde{\pi }}_{2i}}=\frac{{{{\tilde{\pi }}}_{1i}}}{{{\theta }_{0}}}$, and ${{X}^{2}}({{\theta}_{0}})=\frac{{{\left( {{y}_{1i}}-{{n}_{1}}{{{\tilde{\pi }}}_{1i}} \right)}^{2}}} {{{n}_{1}}{{{\tilde{\pi }}}_{1i}}(1-{{{\tilde{\pi }}}_{1i}})}\left\{ 1+\frac{{{n}_{1}}({{\theta}_{0}}-{{{\tilde{\pi }}}_{1i}})}{{{n}_{2}}(1-{\tilde{\pi}_{1i}})} \right\}\).

Noether	\(\hat{\theta}_i\pm z_1-\alpha/2\hat{\sigma}_N\),