05.06.2013 Views

Growth model of the reared sea urchin Paracentrotus ... - SciViews

Growth model of the reared sea urchin Paracentrotus ... - SciViews

Growth model of the reared sea urchin Paracentrotus ... - SciViews

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

some two-dimensional processes like respiration or digestion, i.e.,<br />

exchanges across surfaces) and catabolism (proportional to <strong>the</strong> volume <strong>of</strong><br />

<strong>the</strong> animal, and thus a process quantified in 3D).<br />

The second argument against <strong>the</strong> least-square regression method is its<br />

limitation to <strong>model</strong> a "mean individual" in <strong>the</strong> presence <strong>of</strong> individual<br />

variations in <strong>the</strong> dataset. Yet, <strong>the</strong> concept <strong>of</strong> mean individual lacks<br />

meaning when <strong>the</strong> distribution <strong>of</strong> <strong>the</strong> error is not symmetrical and even,<br />

sometimes, multimodal. Considering a bimodal distribution with two<br />

similar, but well-separated modes, <strong>the</strong> mean value is located just in <strong>the</strong><br />

middle <strong>of</strong> <strong>the</strong> two groups, where <strong>the</strong>re are no individuals! In this example,<br />

<strong>the</strong> median is located at <strong>the</strong> same place and will also be a poor<br />

representation <strong>of</strong> <strong>the</strong> dataset. Quantiles, however, can be used more<br />

efficiently (1 st and 3 rd quartiles will be located around each <strong>of</strong> <strong>the</strong> two<br />

modes). Clearly, a regression that can fit a curve on ano<strong>the</strong>r part <strong>of</strong> <strong>the</strong><br />

distribution than a symmetrical position can be useful in situations where<br />

<strong>the</strong> distribution is asymmetrical or multimodal. For instance when sizebased<br />

intraspecific competition occurs, <strong>the</strong> largest animals are inhibitors,<br />

while <strong>the</strong> smallest ones represent <strong>the</strong> most inhibited fraction. <strong>Growth</strong><br />

curves fitted on large and small individuals thus contain information about<br />

<strong>the</strong> impact <strong>of</strong> <strong>the</strong> interspecific competition (by comparison). This<br />

information is not available using a least-square regression, but it is when<br />

two quantile regressions are used, based on large and small quantiles<br />

respectively.<br />

Quantile regression, as defined by Koenker & Bassett (1978) is an<br />

extension <strong>of</strong> <strong>the</strong> least-absolute deviation regression that fits a function for a<br />

median individual. In quantile regression, <strong>the</strong> objective function (called<br />

here deviance δ1) to be minimized is:<br />

with:<br />

Part IV: A growth <strong>model</strong> with intraspecific competition<br />

n<br />

δ 1= ∑ ρτ( Di−ξ1) (21)<br />

i=<br />

1<br />

146

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!