Models with extremely high SE and errors

questions concerning analysis/theory using program PRESENCE

Models with extremely high SE and errors

Postby Happyme » Mon Feb 24, 2020 1:10 pm

Good morning all,

I have begun to analyze my data collected from four sampling events (Aug, Oct, Mar, Apr). I have 3 site covariates and four sampling covariates. My models are giving me high SE (6+ figures) and some are giving an error (-1.#IND00). I have not had this happen before and am not sure how to interpret the output. The model holding occupancy and detection constant or the models holding one or the other constant are the ones with high SE and the models with site and sample covariates have SE's within a more normal range. I am unsure if my high SE is an indication of the real parameter has approached the upper limit or sparse data. In my first two sampling seasons I did not detect any mudpuppies. However, in the 3rd sampling period (March) I caught mudpuppies at 13 of the 20 sites and then the fourth sampling season I caught mudpuppies at 4 sites. We know these are temperature driven from previous studies. The females in March were all gravid and in the April sampling period I only caught males so the females had laid eggs and were nest guarding. However, it does give my a lot of zeros. I can share my data and models. Any feedback would be greatly appreciated.
Happyme
 
Posts: 15
Joined: Fri Feb 21, 2020 9:06 am

Re: Models with extremely high SE and errors

Postby jhines » Mon Feb 24, 2020 2:12 pm

Hi,

I assume you're running a multi-season model since you mentioned something about occupancy being constant for some models. How many surveys per season?

Since you have no data for the first two seasons, you will be unable to estimate occupancy or detection for them. By modeling detection constant, the model can estimate occupancy, but it will be zero for the 1st two seasons, so I suggest removing the 1st two seasons of data. Leaving them in will artificially inflate the occupancy estimates for seasons 3 and 4.

That leaves you with 20 sites over 2 seasons. I suggest starting with the simplest model and not running models with many covariates. If you have a lot of surveys per season, you might be able to model detection with one or two covariates. I wouldn't try more than one covariate in a model for occupancy. If you get funny standard errors, it usually means the estimated parameter has hit the upper or lower limit, or the model is too complicated for the amount of data.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Models with extremely high SE and errors

Postby Happyme » Mon Feb 24, 2020 3:03 pm

I am using a single-season model. My surveys were within a closed system and met the assumption of no changes to the occupancy status of the site during the sampling period. I held occupancy and detection constant to have a basis to compare various impacts of site and sampling covariates in various models. All four sampling occasions were put into PRESENCE along with site covariates (river morphology as categorical data and cover objects z standardized) and sampling covariates of water temp, pH, conductivity and turbidity. I then ran a separate model to look at trapping method by treatment. These models were used to examine sampling-occasion-specific detection probability estimates for trapping methods. These also had extremely high SE.
Happyme
 
Posts: 15
Joined: Fri Feb 21, 2020 9:06 am

Re: Models with extremely high SE and errors

Postby Happyme » Mon Feb 24, 2020 3:11 pm

These are my models:

λ(.), p(.)
λ(.), p(Water_temp)
λ(.), p(pH)
λ(.), p(Conductivity)
λ(.), p(Turbidity)
λ(River_morph), p(.)
λ(River_morph+Cover_objects), p(.)
λ(Cover_objects), p(.)
λ(Cover_objects), p(Water_temp)
λ(River_morph+Cover_objects), p(Water_temp)
λ(River_morph), p(Water_temp)
λ(Cover_objects), p(Turbidity)
λ(River_morph), p(Turbidity)
Happyme
 
Posts: 15
Joined: Fri Feb 21, 2020 9:06 am

Re: Models with extremely high SE and errors

Postby darryl » Mon Feb 24, 2020 3:19 pm

What are your estimated beta parameter/regression coefficient values? If your large SE is associated with a large estimate (eg >10 or <-10), it's probably a boundary issue (ie real estimates near 0 or 1; which you can easily check). That said, from your description of the data, it doesn't sound like a p(.) model is biologically realistic. Furthermore, 20 isn't a very large sample size, from a statistical perspective) so you should expected to get relatively large SEs

Cheers
Darryl
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Models with extremely high SE and errors

Postby Happyme » Mon Feb 24, 2020 4:25 pm

Thanks so much for offering feedback...

Model AICc deltaAICc Weight #Par
psi(Cover_objects), p(Water_temp) 60.13 0.00 0.6143 3
psi(.), p(Water_temp) 58.63 1.15 0.3456 4
psi(River_morph+Cover_objects), p(Water_temp) 65.59 5.46 0.0401 6

psi(.),p(Water_temp)
estimate std.error
A1 psi.a1 : 28.284905 1080758.104785
B1 P[4].b1 : -4.209683 1.258435
B2 P[1].Water_temp : -4.562392 1.565543


psi(Cover_objects),p(Water_temp)
estimate std.error
A1 psi.a1 : 3.449147 2.807095
A2 psi.Z_CoOb : 3.581500 2.886818
B1 P[4].b1 : -4.261523 1.334626
B2 P[1].Water_temp : -5.028065 1.729431

psi(River_morph+Cover_objects),p(Water_temp)
estimate std.error
A1 psi.a1 : 22.995704 11.181677
A2 psi.Inside : -11.265695 13.233512
A3 psi.Outside : -18.527768 11.295113
A4 psi.Z_CoOb : 10.168413 6.903767
B1 P[4].b1 : -4.263496 1.351677
B2 P[1].Water_temp : -5.218919 1.739687

Estimate of c-hat = 0.5643
Happyme
 
Posts: 15
Joined: Fri Feb 21, 2020 9:06 am

Re: Models with extremely high SE and errors

Postby darryl » Mon Feb 24, 2020 4:45 pm

Looks to be a combination of boundary estimates, small sample size and fitting overly too complex models (for the sample size).

For psi(.) model, the large SE is due to the large estimate, which will give psi = 1 (so at boundary). The large SE is (partially) a consequence of using the logit link for these models.

With a sample size of 20, there's the risk of finding spurious models that fit the data too good. Check the real parameter estimates for psi. If most are either exactly 0 or 1, I'd be pretty cautious about the results. Definitely wouldn't try any models with more than one covariate for psi, and it's even debatable whether you should try any covariates (sorry to tell you).

Have you tried using only the last 2 surveys as Jim suggested? Those first two periods with all 0's may not be helping very much. Going from no detections (twice) to 13 detections in the third survey suggests there's been a big change in detection, assuming they were present the whole time.

Cheers
Darryl
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Models with extremely high SE and errors

Postby Happyme » Mon Feb 24, 2020 6:26 pm

Well... this is depressing lol
For psi(cover_objects), p(water_temp)
DERIVED parameter - Psi-conditional = [Pr(occ | detection history)]

Site psi-cond Std.err 95% conf. interval
psi-cond 1 site1 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 2 site2 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 3 site3 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 4 site4 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 5 site5 : 0.4522 0.4147 0.0301 - 0.9564
psi-cond 6 site6 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 7 site7 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 8 site8 : 0.6812 0.4500 0.0355 - 0.9920
psi-cond 9 site9 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 10 site10 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 11 site11 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 12 site12 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 13 site13 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 14 site14 : 0.7530 0.3591 0.0648 - 0.9926
psi-cond 15 site15 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 16 site16 : 0.9954 0.0254 0.0042 - 1.0000
psi-cond 17 site17 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 18 site18 : 0.3433 0.3883 0.0175 - 0.9386
psi-cond 19 site19 : 0.0663 0.1307 0.0011 - 0.8164
psi-cond 20 site20 : 0.6102 0.4931 0.0262 - 0.9891

for psi(.), p(water_temp)
DERIVED parameter - Psi-conditional = [Pr(occ | detection history)]

Site psi-cond Std.err 95% conf. interval
psi-cond 1 site1 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 2 site2 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 3 site3 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 4 site4 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 5 site5 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 6 site6 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 7 site7 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 8 site8 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 9 site9 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 10 site10 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 11 site11 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 12 site12 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 13 site13 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 14 site14 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 15 site15 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 16 site16 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 17 site17 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 18 site18 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 19 site19 : 1.0000 0.0000 1.0000 - 1.0000
psi-cond 20 site20 : 1.0000 0.0000 1.0000 - 1.0000

Water temp does not look good which is unfortunate. Looking at the data water temp is important in when you can capture mudpuppies. I guess my sample size is not large enough to reflect that. From the literature mudpuppies have not been captured in warmer water temperatures. It is during winter/early spring that captures occur when water temps drop. The last sampling period captured the females laying their eggs and nest guarding event and I only captured males. I did try running the model with only the March and April sampling periods with similar issues. I am at a loss with how to proceed so my statistical analysis and models answer questions pertaining to developing a sampling protocol to survey for mudpuppies in large navigable rivers.
The psi(cover_objects), p(water_temp) model looks better but I am not sure if it is an acceptable fit?
Happyme
 
Posts: 15
Joined: Fri Feb 21, 2020 9:06 am

Re: Models with extremely high SE and errors

Postby darryl » Mon Feb 24, 2020 6:36 pm

These are the psi-conditional estimates, not the psi estimates I was referring to, which are further back up the output.
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Models with extremely high SE and errors

Postby Happyme » Mon Feb 24, 2020 6:46 pm

psi(.), p(water temp)
Site estimate Std.err 95% conf. interval
P[1] 1 site1 : 0.0000 0.0000 0.0000 - 0.0127
P[1] 2 site2 : 0.0000 0.0000 0.0000 - 0.0127
P[1] 3 site3 : 0.0000 0.0000 0.0000 - 0.0126
P[1] 4 site4 : 0.0000 0.0000 0.0000 - 0.0125
P[1] 5 site5 : 0.0000 0.0000 0.0000 - 0.0127
P[1] 6 site6 : 0.0000 0.0000 0.0000 - 0.0125
P[1] 7 site7 : 0.0000 0.0000 0.0000 - 0.0129
P[1] 8 site8 : 0.0000 0.0000 0.0000 - 0.0130
P[1] 9 site9 : 0.0000 0.0000 0.0000 - 0.0123
P[1] 10 site10 : 0.0000 0.0000 0.0000 - 0.0121
P[1] 11 site11 : 0.0000 0.0000 0.0000 - 0.0155
P[1] 12 site12 : 0.0000 0.0000 0.0000 - 0.0154
P[1] 13 site13 : 0.0000 0.0000 0.0000 - 0.0153
P[1] 14 site14 : 0.0000 0.0000 0.0000 - 0.0156
P[1] 15 site15 : 0.0000 0.0001 0.0000 - 0.0166
P[1] 16 site16 : 0.0000 0.0001 0.0000 - 0.0168
P[1] 17 site17 : 0.0000 0.0000 0.0000 - 0.0152
P[1] 18 site18 : 0.0000 0.0001 0.0000 - 0.0166
P[1] 19 site19 : 0.0000 0.0000 0.0000 - 0.0156
P[1] 20 site20 : 0.0000 0.0000 0.0000 - 0.0158
P[2] 1 site1 : 0.1314 0.0612 0.0503 - 0.3020
P[2] 2 site2 : 0.1200 0.0596 0.0432 - 0.2918
P[2] 3 site3 : 0.1147 0.0587 0.0400 - 0.2869
P[2] 4 site4 : 0.1166 0.0590 0.0411 - 0.2886
P[2] 5 site5 : 0.0967 0.0552 0.0301 - 0.2700
P[2] 6 site6 : 0.1152 0.0588 0.0403 - 0.2874
P[2] 7 site7 : 0.1182 0.0593 0.0421 - 0.2901
P[2] 8 site8 : 0.1169 0.0591 0.0413 - 0.2889
P[2] 9 site9 : 0.0135 0.0171 0.0011 - 0.1453
P[2] 10 site10 : 0.0150 0.0185 0.0013 - 0.1501
P[2] 11 site11 : 0.0156 0.0190 0.0014 - 0.1520
P[2] 12 site12 : 0.0155 0.0189 0.0014 - 0.1514
P[2] 13 site13 : 0.0059 0.0092 0.0003 - 0.1130
P[2] 14 site14 : 0.0119 0.0156 0.0009 - 0.1397
P[2] 15 site15 : 0.0078 0.0114 0.0004 - 0.1229
P[2] 16 site16 : 0.0120 0.0158 0.0009 - 0.1404
P[2] 17 site17 : 0.0052 0.0083 0.0002 - 0.1086
P[2] 18 site18 : 0.0058 0.0091 0.0003 - 0.1125
P[2] 19 site19 : 0.0060 0.0093 0.0003 - 0.1134
P[2] 20 site20 : 0.0058 0.0090 0.0003 - 0.1123
P[3] 1 site1 : 0.4793 0.0986 0.2979 - 0.6663
P[3] 2 site2 : 0.4700 0.0968 0.2927 - 0.6551
P[3] 3 site3 : 0.4431 0.0918 0.2774 - 0.6225
P[3] 4 site4 : 0.4709 0.0970 0.2933 - 0.6563
P[3] 5 site5 : 0.4940 0.1015 0.3058 - 0.6839
P[3] 6 site6 : 0.5197 0.1064 0.3193 - 0.7139
P[3] 7 site7 : 0.5551 0.1131 0.3371 - 0.7537
P[3] 8 site8 : 0.5094 0.1045 0.3139 - 0.7020
P[3] 9 site9 : 0.5087 0.1043 0.3136 - 0.7013
P[3] 10 site10 : 0.5764 0.1168 0.3476 - 0.7766
P[3] 11 site11 : 0.5632 0.1145 0.3411 - 0.7625
P[3] 12 site12 : 0.5895 0.1190 0.3539 - 0.7901
P[3] 13 site13 : 0.6248 0.1242 0.3710 - 0.8247
P[3] 14 site14 : 0.2134 0.0673 0.1100 - 0.3732
P[3] 15 site15 : 0.3609 0.0785 0.2247 - 0.5240
P[3] 16 site16 : 0.5598 0.1139 0.3394 - 0.7588
P[3] 17 site17 : 0.5355 0.1095 0.3273 - 0.7319
P[3] 18 site18 : 0.5479 0.1118 0.3335 - 0.7458
P[3] 19 site19 : 0.5483 0.1119 0.3338 - 0.7463
P[3] 20 site20 : 0.5518 0.1125 0.3355 - 0.7501
P[4] 1 site1 : 0.0826 0.0518 0.0231 - 0.2558
P[4] 2 site2 : 0.0946 0.0548 0.0290 - 0.2679
P[4] 3 site3 : 0.0864 0.0528 0.0249 - 0.2597
P[4] 4 site4 : 0.1139 0.0586 0.0396 - 0.2862
P[4] 5 site5 : 0.0783 0.0506 0.0211 - 0.2513
P[4] 6 site6 : 0.0808 0.0513 0.0222 - 0.2539
P[4] 7 site7 : 0.0809 0.0513 0.0223 - 0.2540
P[4] 8 site8 : 0.0821 0.0517 0.0228 - 0.2553
P[4] 9 site9 : 0.2885 0.0713 0.1704 - 0.4447
P[4] 10 site10 : 0.2761 0.0704 0.1604 - 0.4321
P[4] 11 site11 : 0.2690 0.0700 0.1548 - 0.4251
P[4] 12 site12 : 0.2902 0.0714 0.1717 - 0.4464
P[4] 13 site13 : 0.2705 0.0701 0.1560 - 0.4266
P[4] 14 site14 : 0.2691 0.0700 0.1549 - 0.4253
P[4] 15 site15 : 0.2314 0.0681 0.1244 - 0.3895
P[4] 16 site16 : 0.2632 0.0697 0.1501 - 0.4194
P[4] 17 site17 : 0.2528 0.0691 0.1417 - 0.4094
P[4] 18 site18 : 0.2453 0.0688 0.1356 - 0.4024
P[4] 19 site19 : 0.2439 0.0687 0.1345 - 0.4010
P[4] 20 site20 : 0.2286 0.0680 0.1222 - 0.3869

psi(cover objects), p(water temp)
Site estimate Std.err 95% conf. interval
psi 1 site1 : 1.0000 0.0000 0.0002 - 1.0000
psi 2 site2 : 0.9998 0.0013 0.0103 - 1.0000
psi 3 site3 : 0.9946 0.0220 0.0561 - 1.0000
psi 4 site4 : 0.9734 0.0754 0.1082 - 0.9999
psi 5 site5 : 0.7113 0.3001 0.1232 - 0.9774
psi 6 site6 : 0.6769 0.3136 0.1120 - 0.9721
psi 7 site7 : 0.9734 0.0754 0.1082 - 0.9999
psi 8 site8 : 0.8729 0.2056 0.1538 - 0.9962
psi 9 site9 : 0.8168 0.2468 0.1496 - 0.9912
psi 10 site10 : 0.9689 0.0845 0.1142 - 0.9999
psi 11 site11 : 0.9581 0.1039 0.1254 - 0.9997
psi 12 site12 : 0.9720 0.0784 0.1102 - 0.9999
psi 13 site13 : 0.9963 0.0162 0.0471 - 1.0000
psi 14 site14 : 0.8604 0.2159 0.1538 - 0.9952
psi 15 site15 : 0.8729 0.2056 0.1538 - 0.9962
psi 16 site16 : 0.9989 0.0057 0.0255 - 1.0000
psi 17 site17 : 0.4961 0.3650 0.0533 - 0.9451
psi 18 site18 : 0.6769 0.3136 0.1120 - 0.9721
psi 19 site19 : 0.2215 0.3392 0.0060 - 0.9307
psi 20 site20 : 0.8604 0.2159 0.1538 - 0.9952
Happyme
 
Posts: 15
Joined: Fri Feb 21, 2020 9:06 am

Next

Return to analysis help

Who is online

Users browsing this forum: No registered users and 10 guests