Intercept Term

[bgerber]

Sorry for the basic question, but I wanted to make sure I am using PRESENCE correctly.

Does PRESENCE automatically have the constant intercept term in the matrices, or does the 1st column need to be the constant?

Sincerely,
Brian Gerber

[jhines]

No, PRESENCE does not add an intercept term in the design matrix.

Cheers,

Jim

[JT Schloesser]

I have a new question pertaining to the intercept and covariates.

I am modeling occupancy at five habitat types and p with three gear types which are categorical. For background info, some gear types didn't collect a species and some habitats the species was never collected. When I set up my design matrix, I have psi a1=1, a2=hab2, a3=hab3, a4=hab4, and a5=hab5. For p I have p1=1, p2=gear2, p3=gear3. When I run this model the intercept is always 0.5 for p. Why is this and shouldn't it give me a real estimate for p? This concerned me, so what I did was remove the intercept term for psi and p and ran my models as a1=hab1, a2=hab2, a3=hab3, and so on, and for p as p1=gear1, p2=gear2, and p3=gear3. This latter model gives me reasonable estimates for psi and p, even though I don't have an intercept term in there. So my question is why do I get better estimates without an intercept than with one? I am running models for many species and this same pattern is seen among the other species.

Also, if a gear type never caught an animal, I would eliminate that gear type as a covariable, with no intercept and the models would run and give me a reasonable answer. But those samples from the gear type that was ommitted will all be 0.5. With the intercept term in there, my answers would change and give me estimates that didn't seem right to me. I am just confused as to why models without the intercept work better than those with the intercept included?

Thanks for any help!

JTS

[darryl]

I am modeling occupancy at five habitat types and p with three gear types which are categorical. For background info, some gear types didn't collect a species and some habitats the species was never collected. When I set up my design matrix, I have psi a1=1, a2=hab2, a3=hab3, a4=hab4, and a5=hab5. For p I have p1=1, p2=gear2, p3=gear3. When I run this model the intercept is always 0.5 for p. Why is this and shouldn't it give me a real estimate for p? This concerned me, so what I did was remove the intercept term for psi and p and ran my models as a1=hab1, a2=hab2, a3=hab3, and so on, and for p as p1=gear1, p2=gear2, and p3=gear3. This latter model gives me reasonable estimates for psi and p, even though I don't have an intercept term in there. So my question is why do I get better estimates without an intercept than with one? I am running models for many species and this same pattern is seen among the other species.

I presume you actually mean b1, b2 and b3 rather than p1, p2 and p3? And also that your 3 gear types represent your repeat surveys? It sounds like the two models you've fit above are exactly the same model, just a different way of setting it up with the design matrix. You should get the exact same AIC, -2log-like values etc. If not then something is screwy with the way you set up one of the models.

Do you mean the estimated beta-parameter b1 is 0.5 or the real parameter p1 is 0.5?

Also, if a gear type never caught an animal, I would eliminate that gear type as a covariable, with no intercept and the models would run and give me a reasonable answer. But those samples from the gear type that was ommitted will all be 0.5. With the intercept term in there, my answers would change and give me estimates that didn't seem right to me. I am just confused as to why models without the intercept work better than those with the intercept included?

Sounds like you might not be setting up your design matrices quite right. If you post some of the ones that are causing you a problem I'm sure someone can put you on the right track.

Darryl