No significant models

questions concerning analysis/theory using program PRESENCE

No significant models

Postby dmh1454 » Sat Aug 01, 2020 6:05 pm

Hey y'all,

So after running all of my models I generated numerous models within 2 delta qAIC including the null model. However, after calculating the 95% confidence intervals for each model, none are significant as they all cross 0 as seen below. What could be the cause of this? Am I missing a variable that needs added in or does this just mean that no variables are significant?

Parameter β SE LCL UCL
ψ(.)ρ(.)
ψ 0.423539 0.295274 -0.15519804 1.00227604
ψ.Control 1.428572 0.21086 1.0152864 1.8418576

ψ(Aquatic)ρ(.)
ψ 0.43857 0.307246 -0.16363216 1.04077216
ψ.Emergent -0.573449 0.312854 -1.18664284 0.03974484
*ρ 1.428593 0.210853 1.01532112 1.84186488

ψ(Waterway)ρ(.)
ψ 0.917236 0.418589 0.09680156 1.73767044
ψ.Stagnant -1.117416 0.61425 -2.321346 0.086514
*ρ 1.428572 0.21086 1.0152864 1.8418576


As well, I had this problem with another model set. The difference is, the estimates and standard errors are high, causing extremely wide confidence intervals. Only the models above the null model have this problem. I checked over the data and renormalized a lot of it, but I still get the same results for every year I run models with this species.

Any insight would be much appreciated.
Parameter β SE LCL UCL
ψ(Waterway+Herbaceous)ρ(.)
ψ 4.697274 5.860632 -6.789565 16.184113
ψ.Stagnant -6.008358 5.857509 -17.489076 5.472360
ψ.Herbaceous 2.381525 1.364845 -0.293571 5.056621
*ρ -1.235572 0.240584 -1.707117 -0.764027

ψ(Waterway)ρ(.)
ψ 2.625552 2.826669 -2.914719 8.165823
ψ.Stagnant -4.082469 2.791899 -9.554591 1.389653
*ρ -1.205787 0.309348 -1.812109 -0.599465

ψ(Stream Density)ρ(.)
ψ 4.576777 5.018415 -5.259316 14.412870
ψ.Stream 7.258050 6.188546 -4.871500 19.387600
*ρ -1.362518 0.285677 -1.922445 -0.802591
dmh1454
 
Posts: 11
Joined: Fri Aug 02, 2019 12:35 pm

Re: No significant models

Postby jhines » Mon Aug 03, 2020 7:38 am

It sounds like know that some variable ought to explain the variation in psi. Unless you simulated the data, you’ll never know for sure. It’s possible that occupancy is simply random or depends on a covariate that you didn’t measure. Another possibility is that there isn’t enough data to model the relationship accurately.
However, the 2nd and 3rd models are “significant” at the 90% level. The definition of significance is arbitrary, and the most commonly used value is 95%. I think the more important question is not simply is the relationship significant, but does the covariate affect the parameter in the expected direction and how “strong” is the relationship (ie. how much different from no relationship)?
For the 2nd model, the estimate indicates that there is a negative relationship between occupancy and the emergent covariate. The probability of getting a beta value <= -0.57 is less than .05. If you hypothesized that the relationship should be negative, then you can use a one-tailed test and this result is “significant” at the 95% level.
The 3rd model is similar to the 2nd. If the direction of the effect is what you hypothesized, then you can use a one-tailed test and the p-value < .05 and you can call it “significant”.
Note: It appears the 1st model is mis-labelled and “psi.control” should be “*p”. If those are the labels of the beta estimates from PRESENCE, then I suspect that your design matrices have a problem.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: No significant models

Postby dmh1454 » Tue Aug 04, 2020 6:40 pm

Thank you for the help.

Follow up question; the post doc assisting me wishes for me to check to see if the predictions are reasonable. I was unable to find anything in the forum or through the PRESENCE help files but noticed you spoke about simulating the data. Are those the same thing or must I run predictions through R?
dmh1454
 
Posts: 11
Joined: Fri Aug 02, 2019 12:35 pm

Re: No significant models

Postby jhines » Tue Aug 04, 2020 8:24 pm

I'm not sure what you mean by predictions being reasonable. You can simulate data using program GENPRES which is included with PRESENCE (look in PRESENCE group in start menu).
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA


Return to analysis help

Who is online

Users browsing this forum: No registered users and 11 guests

cron