02.09.2014 Views

multivariate poisson hidden markov models for analysis of spatial ...

multivariate poisson hidden markov models for analysis of spatial ...

multivariate poisson hidden markov models for analysis of spatial ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

ecause the <strong>for</strong>mer captures more <strong>of</strong> the existing variance in the data. However, from a<br />

practical aspect, the model with common covariance structure is preferable to the fully<br />

structured model because it requires fewer parameters to be estimated.<br />

There<strong>for</strong>e, the important question is whether a model somewhere in between the two<br />

presented extreme <strong>models</strong> can be found, that is, both a) theoretically good enough to<br />

describe most <strong>of</strong> the existing covariances, and b) practically suitable in terms <strong>of</strong> the<br />

number <strong>of</strong> parameters to be estimated.<br />

The model introduced in this section was tried to address the above mentioned problem.<br />

The main idea is to simplify the variance/covariance structure as much as possible by<br />

including only statistically significant m-fold interactions. For this trivariate model, the<br />

statistical significance <strong>of</strong> the weed count interactions between Wild Buckwheat ( Y 1<br />

),<br />

Dandelion ( Y 2<br />

) and Wild Oats ( Y 3<br />

) will study by using <strong>of</strong> loglinear <strong>analysis</strong> (see section<br />

5.7 and section 6.3). The loglinear <strong>analysis</strong> is particularly appropriate <strong>for</strong> the<br />

development <strong>of</strong> a simpler <strong>multivariate</strong> Poisson mixture model <strong>for</strong> clustering since it<br />

facilitates to discover, which interaction terms in the variance/covariance matrix can be<br />

set equal to zero. We will show that (section 6.3) the p values <strong>of</strong> the goodness <strong>of</strong> fit<br />

statistics <strong>of</strong> the <strong>models</strong> with some <strong>of</strong> the two-fold interactions and without any<br />

interaction do not differ very much. There<strong>for</strong>e, the two-fold interactions were kept in<br />

the model. The latent variables, X = X , X , X , X , X , ) and the vector <strong>of</strong><br />

(<br />

1 2 3 12 13<br />

X<br />

23<br />

parameters, θ = ( θ1, θ2, θ3, θ12, θ13, θ23)<br />

, are used to present the model and thus the<br />

restricted covariance model can be defined as:<br />

67

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

Saved successfully!

Ooh no, something went wrong!