|
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
If you reply to this message, it will go to all FISH-SCI members.
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
Hello Fishsci list.
I fully agree with the use of GLM in this analysis and expected this to be the response from the initial posting. I use SPSS for GLMs because it can generate a fully factored model with the click of an option button without all the tedious typing of the model necessary in MINITAB (for example). I've not had any experience of Program R though would welcome news, please.
I'm definitely no salesman for SPSS but it can also cope with more than one dependent variable (multivariate GLM). Do be cautious however, that the data (the dependent variable,s) approximates a normal distribution (check it), though you can transform it or apply a weighted least squares weight which might help, though won't work miracles.
Only factors (categories) are compared as interactions (a, b, a*b etc) so it may be better having the data in this form rather than as co-variables, though you might loose precision. SPSS has a bewildering array of outputs but an important test is the homogeneity test (Levene's test of equality of error variance). If P < 0.05 then the model is invalid and you'll have to reassess your list of factors and co-variables. GLM is a parametric test and the usual assumptions of these must be employed.
Post hoc testing should be done to check the form that significant interactions take. I usually do this by a suitable test between the dependent variable and the factor (anova etc) or co-variable (regression) responsible. These post hoc test have their own assumptions which you must note, too.
Finally:
GLM is a very powerful tool which can cope with a wide variety of data types, though the initial model must be valid.
Stephen Cotterell
-Stephen Cotterell
-Rm 4, 22 Portland Square
-School of Earth, Ocean and Environmental Science
-University of Plymouth
-Drake Circus
-PLYMOUTH
-PL4 8AA
-T: +44 (0) 1752 232411
-F: +44 (0) 1752 232406
-E: [log in to unmask]
-----Original Message-----
From: Scientific forum on fish and fisheries
[mailto:[log in to unmask]]On Behalf Of Antonio Olinto
Sent: 22 June 2004 12:52
To: [log in to unmask]
Subject: Re: fisheries data analysis
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
If you reply to this message, it will go to all FISH-SCI members.
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
Dear Abad,
Another statistic tool to access the magnitude of the effects of your variables
is the analysis of deviance, given in the generalized linear models. In GLM you
can use both category and continuous data, and also you can choose
distributions other than gaussian, like gamma.
There's an example in Quinn & Deriso 1999 (Quantitative Fish Dymamics). See
pages 18 to 25. They check the effects of year, area and depth on CPUE. You can
do the same with shell height. In the example they use normal distribution, log
link function and all variables are categories. Depending on your data, you can
improve this.
I use the program R (www.r-project.org), it's a freeware and runs on different
plataforms. If you want I can send you a pdf demostrating how to solve Quinn &
Deriso's example in R.
Best regards,
Antonio Olinto Avila da Silva
Sao Paulo Fisheries Institute
Brazil
Citando Antonio Hervas Abad <[log in to unmask]>:
> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
> If you reply to this message, it will go to all FISH-SCI members.
> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
>
> I'm working with scallop (pecten maximus) stock assessment in Ireland. One
> of the stock assessment procedures that i am carrying on is a sampling
> programm from the landings. Shell height is recoreded from each landing.
> After several measurements from differents vessels, differents fishing
> areas and different time of the year i want to study the effect of vessels,
> season and location, on the scallop size structure obtained from the
> sampling. I understand that analysis of variance (ANOVA or MANOVA) is the
> method to use for this study but maybe some of you has experience working
> with similars studies and can advice to me of which statistic analysis is
> the more suitable.
>
> Thanks very much
>
> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
> To leave the Fish-Sci list, Send blank message to:
> mailto:[log in to unmask]
> For information send INFO FISH-SCI to [log in to unmask]
> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
-------------------------------------------------
WebMail Bignet - O seu provedor do litoral
www.bignet.com.br
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
To leave the Fish-Sci list, Send blank message to:
mailto:[log in to unmask]
For information send INFO FISH-SCI to [log in to unmask]
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
To leave the Fish-Sci list, Send blank message to:
mailto:[log in to unmask]
For information send INFO FISH-SCI to [log in to unmask]
><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><> ><>
|
