occuMulti

102 views
Skip to first unread message

Ye Htet Lwin

unread,
Dec 23, 2021, 10:29:41 AM12/23/21
to unmarked

Dear unmarked users, 

I need some guidance to explain the confusion result for my separate models. I build two different models which include four species in each model and two species are common (i.e., NRM and MS) in both models.

m1=NRM+MS+GM+RS

m2=NRM+MS+WB+SB

When I run the first model, m1, I got interaction result between NRM and MS with negative sign, but it was changed into positive sign in m2. My view is that same species pair from identical dataset should has same interaction effect (i.e., positive and negative) whatever they are combining with other species. Based on my understanding, my questions are -

1.     Why sign in interaction term is difference between model 1 and 2 since they are same species (i.e., NRM and MS pair) in m1 and m2?

2.     If it is usual case, what are the driving factors for that?

 Thanks in advance for any help.

Lwin

 ###m1=NRM+MS+GM+RS

 Call:

occuMulti(detformulas = c("~d1+d2", "~d1", "~d1+d2", 

    "~1"), stateformulas = c("~s22+s4+s5+s6+s7", 

    "~s2+s22+s3+s4", "~s2+s22+s4", "~s22+s4+s6+s7", 

    "~1", "~1", "~1", "~1", "~1", 

    "~1", "~0", "~0", "~0", "~0", 

    "~0"), data = umf_multi)

 

Occupancy:

                      Estimate    SE      z  P(>|z|)

[NRM] (Intercept)        1.122 0.562  1.996 4.59e-02

[NRM] s22               -2.287 0.669 -3.421 6.24e-04

[NRM] s4                -0.660 0.327 -2.020 4.34e-02

[NRM] s5                -1.098 0.575 -1.909 5.63e-02

[NRM] s6                 0.453 0.403  1.125 2.60e-01

[NRM] s7PAs              1.304 0.897  1.454 1.46e-01

[MS] (Intercept)       -0.198 0.566 -0.349 7.27e-01

[MS] s2                 5.342 1.867  2.861 4.22e-03

[MS] s22               -5.606 1.830 -3.064 2.18e-03

[MS] s3                 0.519 0.237  2.192 2.84e-02

[MS] s4                 0.946 0.316  2.996 2.74e-03

[GM] (Intercept)       -3.895 1.000 -3.896 9.80e-05

[GM] s2                12.653 3.232  3.915 9.02e-05

[GM] s22              -11.229 2.974 -3.776 1.59e-04

[GM] s4                 0.991 0.394  2.518 1.18e-02

[RS] (Intercept)       -4.308 1.208 -3.567 3.61e-04

[RS] s22               -0.907 0.520 -1.744 8.12e-02

[RS] s4                 0.362 0.407  0.890 3.74e-01

[RS] s6                 0.683 0.376  1.819 6.90e-02

[RS] s7PAs              2.124 1.035  2.053 4.01e-02

[NRM:MS] (Intercept)   -0.511 0.662 -0.772 4.40e-01

[NRM:GM] (Intercept)    2.258 0.884  2.556 1.06e-02

[NRM: RS] (Intercept)    0.521 0.967  0.539 5.90e-01

[MS:GM] (Intercept)    0.346 0.719  0.481 6.30e-01

[MS: RS] (Intercept)    1.483 0.691  2.147 3.18e-02

[GM: RS] (Intercept)    0.262 0.733  0.357 7.21e-01

 

Detection:

                  Estimate     SE       z  P(>|z|)

[NRM] (Intercept)  -0.5387 0.0711  -7.578 3.51e-14

[NRM] d1L          -0.8683 0.1452  -5.980 2.23e-09

[NRM] d1M          -0.2216 0.0990  -2.240 2.51e-02

[NRM] d2           -0.0879 0.0485  -1.812 6.99e-02

[MS] (Intercept)  -1.1845 0.1267  -9.350 8.73e-21

[MS] d1L          -0.6396 0.2322  -2.755 5.87e-03

[MS] d1M          -1.4321 0.2313  -6.192 5.93e-10

[GM] (Intercept)  -2.9177 0.2855 -10.218 1.64e-24

[GM] d1L           0.1956 0.4800   0.407 6.84e-01

[GM] d1M           1.1800 0.3169   3.723 1.97e-04

[GM] d2            0.2283 0.1297   1.760 7.83e-02

[RS] (Intercept)  -2.4859 0.2276 -10.921 9.19e-28

 

AIC: 4977.662

###m2=NRM+MS+WB+SB###################

Call:

occuMulti(detformulas = c("~d1+d2", "~d1", "~d2", 

    "~d1+d2"), stateformulas = c("~s22+s4+s5+s6+s7", 

    "~s2+s22+s3+s4", "~s22+s7", "~s22+s3", 

    "~1", "~1", "~1", "~1", "~1", 

    "~1", "~0", "~0", "~0", "~0", 

    "~0"), data = umf_multi)

 

Occupancy:

                      Estimate    SE       z  P(>|z|)

[NRM] (Intercept)       0.9730 0.570  1.7083 8.76e-02

[NRM] s22              -2.1041 0.637 -3.3037 9.54e-04

[NRM] s4               -0.6802 0.312 -2.1793 2.93e-02

[NRM] s5               -0.9763 0.496 -1.9678 4.91e-02

[NRM] s6                0.5737 0.380  1.5114 1.31e-01

[NRM] s7PAs             1.6323 0.887  1.8395 6.58e-02

[MS] (Intercept)      -0.6003 0.569 -1.0549 2.91e-01

[MS] s2                6.6717 1.672  3.9891 6.63e-05

[MS] s22              -6.6706 1.664 -4.0096 6.08e-05

[MS] s3                0.5423 0.253  2.1417 3.22e-02

[MS] s4                1.1317 0.325  3.4781 5.05e-04

[WB] (Intercept)      -1.9355 0.833 -2.3229 2.02e-02

[WB] s22              -1.4748 0.402 -3.6677 2.45e-04

[WB] s7PAs             1.3911 0.634  2.1940 2.82e-02

[SB] (Intercept)      -3.6689 1.299 -2.8242 4.74e-03

[SB] s22              -3.5651 1.122 -3.1781 1.48e-03

[SB] s3                0.7278 0.361  2.0136 4.40e-02

[NRM: MS] (Intercept)   0.1209 0.590  0.2050 8.38e-01

[NRM:WB] (Intercept)   0.7037 0.736  0.9568 3.39e-01

[NRM:SB] (Intercept)   0.3864 0.936  0.4128 6.80e-01

[MS:WB] (Intercept)   0.5932 0.557  1.0643 2.87e-01

[MS:SB] (Intercept)   0.2054 0.698  0.2942 7.69e-01

[WB:SB] (Intercept)   0.0288 0.883  0.0326 9.74e-01

 

Detection:

                  Estimate     SE       z  P(>|z|)

[NRM] (Intercept)   -0.539 0.0711  -7.577 3.54e-14

[NRM] d1L           -0.877 0.1458  -6.011 1.84e-09

[NRM] d1M           -0.224 0.0991  -2.258 2.40e-02

[NRM] d2            -0.089 0.0486  -1.830 6.72e-02

[MS] (Intercept)   -1.192 0.1278  -9.327 1.09e-20

[MS] d1L           -0.663 0.2341  -2.830 4.65e-03

[MS] d1M           -1.445 0.2308  -6.259 3.86e-10

[WB] (Intercept)   -2.291 0.1311 -17.472 2.36e-68

[WB] d2            -0.332 0.1316  -2.527 1.15e-02

[SB] (Intercept)   -2.559 0.4297  -5.955 2.59e-09

[SB] d1L           -0.119 0.7441  -0.160 8.73e-01

[SB] d1M            1.119 0.4554   2.456 1.40e-02

[SB] d2            -0.154 0.8404  -0.184 8.54e-01

 

AIC: 5115.912

 

Ken Kellner

unread,
Dec 23, 2021, 11:40:17 AM12/23/21
to unmarked
Hi Lwin,

You have quite a lot of covariates in your model. It is possible that the covariates included in the first-order terms could affect the sign of the interaction terms. It's pretty hard to understand what specifically might be going on with such a complex model. I'd suggest comparing a series of simpler models (perhaps starting with just one covariate) and seeing how the interaction(s) changes in each simpler case.

In this specific pair of models, though, I would say that for both models there is no evidence of an interaction in either direction between NRM and MS. The signs of the estimates may be different, but the corresponding SEs are very large and p >>> 0.05 in both cases.

Ken

Ye Htet Lwin

unread,
Dec 23, 2021, 11:58:12 AM12/23/21
to unma...@googlegroups.com
Hello Ken,
Thank you very much for your prompt response sir. Actually, I took these several variables from the best fit model of single season occupancy's result because I am stuck with choosing proper covariates for co-occurrence models after running a single season model. So, could you please give me any suggestion to decide how to consider the covariates for co-occurrence models after running a single season model. 
Thanks for your valuable answer.
Lwin

--
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/2ba1f64b-1d87-4904-8f56-8606e0a900acn%40googlegroups.com.

Ken Kellner

unread,
Dec 23, 2021, 12:13:51 PM12/23/21
to unmarked
Hi Lwin,

Really the answer depends on your goals and hypotheses. I think it's reasonable to use results from single-species models as a starting point when setting up multispecies models. I didn't mean to imply that you shouldn't use the model that you sent earlier, just that you might want to also consider some simpler models to see how things change and perhaps get a better understanding of the complex model.

Ken

Reply all
Reply to author
Forward
0 new messages