glm) with link
function instead of simple quadratic curve. The test was popularized
in ecology for the analysis of humped species richness patterns
(Mittelbach et al. 2001), but it is more general. With logarithmic
link function, the quadratic response defines the Gaussian response
model of ecological gradients (ter Braak & Looman 1986), and the test
can be used for inspecting the location of Gaussian optimum within a
given range of the gradient. It can also be used to replace Tokeshi's
test of MOStest(x, y, interval, ...)
## S3 method for class 'MOStest':
plot(x, which = c(1,2,3,6), ...)
fieller.MOStest(object, level = 0.95)
## S3 method for class 'MOStest':
profile(fitted, alpha = 0.01, maxsteps = 10, del = zmax/5, ...)
## S3 method for class 'MOStest':
confint(object, parm = 1, level = 0.95, ...)plot.x are used.which=1 and
2 define plots specific to MOStest (see Details), and
larger values select graphs of plot.lm (minus 2).MOStest.glm, and it returns the result
of object of glm amended with the result of the test. The new
items in the MOStest are:TRUE if the response is a
hump.TRUE if the hump or the pit is bracketed by
the evaluated points.family and link function. If $b_2
< 0$, this defines a unimodal curve with highest point at $u =
-b_1/(2 b_2)$ (ter Braak & Looman 1986). If $b_2 > 0$, the
parabola has a minimum at $u$ and the response is sometimes
called x is shifted to the values $p_1$ and
$p_2$, and the test statistic is based on the differences of
deviances between the original model and model where the origin is
forced to the given location using the standard
anova.glm function (Oksanen et al. 2001).
Mitchell-Olds & Shaw (1987) used the first degree coefficient with
its significance as estimated by the summary.glm
function. This give identical results with Normal error, but for
other error distributions it is preferable to use the test based on
differences in deviances in fitted models.The test is often presented as a general test for the location of the hump, but it really is dependent on the quadratic fitted curve. If the hump is of different form than quadratic, the test may be insignificant.
Because of strong assumptions in the test, you should use the support
functions to inspect the fit. Function plot(..., which=1)
displays the data points, fitted quadratic model, and its approximate
95% confidence intervals (2 times SE). Function plot with
which = 2 displays the approximate confidence interval of
the polynomial coefficients, together with two lines indicating the
combinations of the coefficients that produce the evaluated points of
x. Moreover, the cross-hair shows the approximate confidence
intervals for the polynomial coefficients ignoring their
correlations. Higher values of which produce corresponding
graphs from plot.lm. That is, you must add 2 to the
value of which in plot.lm.
Function fieller.MOStest approximates the confidence limits
of the location of the extreme point (hump or pit) using Fieller's
theorem following ter Braak & Looman (1986). The test is based on
quasideviance except if the family is poisson
or binomial. Function profile evaluates the profile
deviance of the fitted model, and confint finds the profile
based confidence limits following Oksanen et al. (2001).
The test is typically used in assessing the significance of diversity hump against productivity gradient (Mittelbach et al. 2001). It also can be used for the location of the pit (deepest points) instead of the Tokeshi test. Further, it can be used to test the location of the the Gaussian optimum in ecological gradient analysis (ter Braak & Looman 1986, Oksanen et al. 2001).
Mittelbach, G.C. Steiner, C.F., Scheiner, S.M., Gross, K.L., Reynolds, H.L., Waide, R.B., Willig, R.M., Dodson, S.I. & Gough, L. 2001. What is the observed relationship between species richness and productivity? Ecology 82, 2381--2396.
Oksanen, J., Läärä, E., Tolonen, K. & Warner, B.G. 2001. Confidence intervals for the optimum in the Gaussian response function. Ecology 82, 1191--1197.
ter Braak, C.J.F & Looman, C.W.N 1986. Weighted averaging, logistic regression and the Gaussian response model. Vegetatio 65, 3--11.
humpfit.## The Al-Mufti data analysed in humpfit():
mass <- c(140,230,310,310,400,510,610,670,860,900,1050,1160,1900,2480)
spno <- c(1, 4, 3, 9, 18, 30, 20, 14, 3, 2, 3, 2, 5, 2)
mod <- MOStest(mass, spno)
## Insignificant
mod
## ... but inadequate shape of the curve
op <- par(mfrow=c(2,2), mar=c(4,4,1,1)+.1)
plot(mod)
## Looks rather like log-link with Poisson error and logarithmic biomass
mod <- MOStest(log(mass), spno, family=quasipoisson)
mod
plot(mod)
par(op)
## Confidence Limits
fieller.MOStest(mod)
confint(mod)
plot(profile(mod))Run the code above in your browser using DataLab