Taylor’s Open Science Product

I am following Amanda’s good example and posting links to my open science products. I am finding Piktochart isn’t really great at sharing unless you have the pro version.

This is my infographic for my systematic reivew:
https://magic.piktochart.com/output/3234414-stats-in-desert-restoration

This is the addition to my infographic. It’s an interesting tidbit about Nathan Mantel, the statistician who created the Mantel Test:
https://magic.piktochart.com/output/3233790-mantel-man-of-mystery

I also put my systematic review dataset on figshare if anyone is interested:
http://figshare.com/articles/Systematic_review_of_the_use_of_statistics_in_restoration_ecology_of_arid_areas_/1200152

GLM study notes

1) what is the difference between a GLM and a GLMM?

GLM: only has fixed effects
GLMM: has fixed and random effects

both have: a set distribution and associated link function,  response variable not assumed to be linear or normal, response variable should be continuous, explanatory variables are categorical and can have continuous covariates

2)What is the difference between fixed and random effects? 

Fixed: this is the treatment in your experiment, should have an effect on your response (slope changes)
Ex: soil characteristics are the fixed effect when testing the influence of soil on plant growth of different species

Random: intercept changes but should not have an effect on your response (slope does not change)
Ex: species is a random effect when testing the influence of soil characteristics on plant growth of different species

3) Checklist of steps for running a GLM/GLMM (Box 4 Bolker)

  • specify model: define – fixed effects, random effects, covariates, response variable, a priori  decide on the number of factors to include in the model
  • choose distribution and appropriate link function
  • check assumptions: homogeniety of variances, outliers, does your distribution match the assumed distribution? (goodness of fit statistics). If variances are not homogeneous or the distribution does not match then change your model, adding effects or covariates, interactions etc
  • recheck assumptions of final model (see previous step)
  • check overdistribution (especially for poisson distributions)
  • can choose/compare models using AIC values (lower AIC = better)

4) Common errors that are made when performing a GLM/GLMM

Selecting the wrong distribution and/or the wrong link function
not checking the fit of the model to the distribution

 

Ordination!

Hi everyone!

Here is a paper that has a nice summary of different ordination methods and how they’re used. It touches on CCA and RDA. It’s pretty math-y but we won’t be going into that much detail, especially since not everyone will be using these stats in their field.

-Katrina & Sam

Peres-Neto et al. 2006

CART!

Hi everyone!

Please find attached a paper by De’ath and Fabricus (2000) on CART (Classification and Regression trees). I think this paper really outlines what CART is and how it can be used.  I suggest everyone read it before thursdays class so that you get an understanding of what we will be teaching.

A paper on ordinations is coming soon!

-Sam & Katrina

De’ath and Fabricus 2000

unofficial blog for biostats