# Significant presence/absence of plant species [closed]

I have approximately 300 cluster of 200 different plant species and a presence absence matrix of all species versus clusters (since some species don't have clusters so they are checked if they are present or absent in other clusters). Now I want to find out if there is significant presences or absences of this plants in their clusters. At the same time I need to know which clusters have this significant presences. I want to know that if there is spesific software can do this or do I have to write code?

edit: it is not clear what i mean with significant presence and absence as i understand. by significant presence i mean "species being in the same clusters all the time. not one of them being in another cluster without the other." by significant absence i mean that "if leys say Y species never tooks place in clusters of which X species exist."

• Exactly what do you mean by "significant presences or absences"? – fileunderwater Dec 19 '16 at 13:13
• significant presence means: two (or more) species being in always together in clusters. significant absence means: lets say X species never tooks place in a cluster if Y species exist in it. – user28219 Dec 20 '16 at 7:28
• Are you with "clusters" referring to plant presences at different sites (i.e. a species co-occurrence matrix) or something else? When you say "clusters" I mostly think about groups created by some sort of algorithm (i.e. clusters) to find patterns in data. – fileunderwater Dec 20 '16 at 8:23

I am not quite sure what you mean by "significant presence". If you observe any non-zero number of plants of a given species in a cluster, then the rate at which this plant occur in this cluster is different from zero. No stats required to infer such thing. You should clarify your post.

Logistic Regression

It sounds like you would like to do a logistic regression with presence / absence as response and the nominal variable cluster as solo explanatory variable. In R you could for example do something like

# Dummy data
d = data.frame(
presence = sample(0:1, 120, replace=TRUE),
cluster =  rep(LETTERS[1:4], 30)
)

# Logistic Regression
m = glm(data= d, formula = presence ~ cluster, family = binomial(link="logit"))
summary(m)


You can find here another example of logistic regression in R

Phylogenetic signal and pseudoreplication

You might want to make sure to decorrelate your variable from the phylogeny to avoid pseudoreplication. There, the solution you might end up going with will depend on your genetic data. If you have a genetic PCA, you could go with using the first 2 axes as simple covariates in a type I sum of squares kind of model (which is probably the default of the logistic regression I performed above). If you have a kinship matrix you could use it with lmekin or MCMCglmm (see here for adding a kinship matrix). There would have many different statistical tools for such decorrelation depending on your data.

• By "significant presence in clusters" the probability of lets say 3 species being together in one cluster is not due to chance. They are being together in every cluster, inot one of them present in another cluster. – user28219 Dec 20 '16 at 7:25
• It phrasing "Is a probability due to chance?" makes no sense. You can only ask the probability of something to happen under a specific model. The expression "significant presence in clusters" is still unclear. You might want to follow an intro course to statistics to understand the logic and context of probabilities and statistical testing. – Remi.b Dec 20 '16 at 7:48
• i think it is pretty clear what i am asking. i have clusters of plants. IF ONE SPECIES IS ALWAYS WITH OTHER SPECIES IN ALL CLUSTERS, it means significant presence. It means that one of the species can not exist without the other. i really don't understand which part of this unclear. – user28219 Dec 20 '16 at 7:53
• Or, one species can not be together with one other spesific species, it means significasnt absence. It means they can not live together, and that's why they are never located in the same clusters. – user28219 Dec 20 '16 at 7:55
• Well this sounds very different from what you said before. It just mean that you want to compute interaction (probably all two-way interactions only) in a logistic regression. – Remi.b Dec 20 '16 at 8:53