SE of zero

[David]

Hello all,
I am trying to construct occupancy models. I have presence/absence data and a few environmental variables. I am getting estimated PSI of 1 (and of course SE of zero) for some of the models. I am not sure why.
Your help is most welcome.

David

[dhewitt]

Do all of your detection histories include at least one '1'?

If not, how many detection histories do you have?

[David]

i have sampled 45 sites. i have visited each site 3 times. in 26 sites i had no detection in any of the visits.

[dhewitt]

Can you paste in your encounter histories? It sounds like an estimation problem, but we need to know a little more about the models you're trying to fit and the encounter histories with detections. In a simple single-season model with no covariates, surely you get a "decent" estimate?

[David]

the encounter history:
100
0..
100
111
11
0..
111
10
11.
0..
0..
0..
0..
1..
0..
0..
0.11
0..
100
0..
0..
0..
0..
0..
..0
0..
0..
0..
0..
.0.
0..
1..
0..
.0.
0..
100
0..
100
101
111
0..
11.
11.

[bacollier]

David,
Other than a couple of edits to your inp file (row 5 has 2 values, so I made the 3rd occasion a "."; row 17 reads 4 occasions, i deleted the last "1"), it seemed to work ok for me in MARK using a standard model? What model are you trying to run run; did it include covariates (& where are they in your enc hist below?)


LOGIT Link Function Parameters of {Psi(.), p(.)}
95% Confidence Interval
Parameter Beta Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p -0.1670541 0.4096732 -0.9700136 0.6359054
2:Psi 1.8368799 2.1191981 -2.3167484 5.9905081


Real Function Parameters of {Psi(.), p(.)}
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p 0.4583333 0.1017071 0.2748778 0.6538273
2:Psi 0.8625793 0.2512018 0.0897453 0.9975039

[David]

hi,
i am using 8 covariates. i am trying different models. for example, if i try 'global model' that includes all 8 var {p(.)psi(8var)}, the result is (i am getting this result also for other cavariates combinations (e.g. {p(.)psi(var2,var3,var4,var5,var6,var7,var8)}):

95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p 0.5000000 0.0000000 0.5000000 0.5000000
2:Psi 1.0000000 0.0000000 1.0000000 1.0000000

there are other models that the result is different ({p(.)psi(var5,var6,var7,var8)}

95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p 0.5000000 0.0000000 0.5000000 0.5000000
2:Psi 0.5446533 0.1435069 0.2778706 0.7880531

this is the input file-
100 1 151.1621704 22.36067963 1525.516357 2177.062256 209 14.62057304 77.82855225 368;
000 1 155.241745 85.44004059 1420.880005 2405.311768 172 19.12505341 28.41265869 398;
100 1 146.0308228 710.2816162 264.6223755 18.02775574 279 1.811248064 108.4349518 99;
111 1 754.204895 560 1068.000488 20.61552811 448 15.020257 63.43493652 891;
11. 1 117.1537476 76.1577301 1023.095764 236.9071503 469 10.7797718 113.1985931 891;
000 1 2198.010498 3028.531006 1708.105713 2107.729736 300 11.64863491 165.9637604 1297;
111 1 940.850708 766.5507202 1656.751343 304.384613 320 5.389322281 122.0053864 236;
10. 1 1940.057983 1424.429688 563.0719604 148.4082184 451 10.05636406 49.5739212 1468;
11. 1 302.3243408 44.72135925 794.8899536 367.1511841 441 8.293805122 5.90612793 1506;
00. 1 165.0757446 1175.967651 789.461853 0 261 6.56633997 124.3803482 2316;
0.. 1 195 1357.681885 1708.581299 28.28427124 238 7.662350178 41.98721314 1291;
000 1 768.6514282 564.3580322 2923.978516 0 251 2.180704594 156.8014069 1144;
00. 1 199.624649 455.4118958 25 0 40 16.89062691 72.75854492 2008;
1.. 1 173.5655518 0 4125.303223 4871.69873 238 15.49434853 97.25320435 1244;
000 1 1108.659058 40 507.7400818 618.9709473 150 9.664408684 139.7636414 316;
000 1 791.5964966 50 661.4000244 535.7704468 158 0.405135602 45 316;
.11 1 359.2005005 545.7105713 289.9137878 0 318 10.92391682 163.4429321 183;
0.. 1 415.4816589 450 1847.220825 409.4203186 370 5.710593224 180 399;
100 1 914.2483521 942.6027832 1445.345947 980.4591064 320 6.000835419 25.34616089 187;
000 1 25 852.7602539 908.3088379 87.32124329 239 8.483933449 13.57043457 18;
000 1 348.8911133 192.0937347 355.1408081 553.2178345 428 10.81600571 42.87890625 27;
00. 1 1228.678101 100.4987564 238.3799438 472.8900452 163 12.85772896 118.810791 424;
000 1 533.8773193 445.5333862 1520.296021 125.2996445 154 14.43409443 65.92449951 1825;
000 1 1198.801514 1860.967529 1869.525391 0 330 8.815243721 110.7722473 1457;
..0 1 574.0426636 329.8484497 731.7445068 443.0011292 425 11.32369614 2.86239624 49;
000 1 103.0776443 647.7653809 10 241.8677368 35 15.27659416 66.2505188 3;
000 1 151.8222656 90 3404.563232 4582.450195 177 23.43271255 11.30993271 268;
000 1 1186.518066 581.8075562 248.6463318 834.2960815 60 16.25391769 95.90612793 60;
000 1 180.3468933 197.9898987 2487.810303 319.4526672 370 19.29004669 36.86989594 1030;
.0. 1 634.0544434 382.7531738 1015.923218 40 448 5.667882919 49.08560181 868;
000 1 84.85281372 1026.255371 973.0621948 735.0680542 400 9.064049721 32.19573212 1381;
01. 1 1350.148193 2010.024902 2394.895508 1366.107666 300 0 0 512;
0.. 1 150 829.2767944 224.1093445 211.4828644 510 8.455702782 160.3461761 63;
.0. 1 77.6208725 834.0863037 1067.824951 196.4688263 380 4.731286049 115.0168915 1072;
000 1 70.7106781 1650.96936 2491.605957 159.7654572 355 6.651732922 120.9637604 1457;
100 1 105.1189804 676.0177612 120.8304596 594.7478638 468 4.222151279 118.3007507 277;
000 1 775 759.276001 26.92582321 70.7106781 42 17.5236721 10.95407105 2763;
100 1 1044.856445 898.8881836 2054.221191 50 248 1.620137453 45 1144;
101 1 961.0671387 1006.479004 1262.457886 733.7744751 316 8.697149277 78.69006348 619;
111 1 1599.976563 1464.923218 2470.951172 1624.81543 330 11.58847809 88.60281372 65;
000 1 100 410.4875183 505.8902893 255.1959991 119 0.905850351 18.43493652 1040;
011 1 989.2547607 790.0632935 1908.487671 60.4152298 300 8.138319969 53.53076172 58;
11. 1 14.14213562 957.5489502 229.4013977 315.0396729 490 4.064796925 50.71059418 7;


thanks

[dhewitt]

David,

Being blunt, you have no chance of estimating effects for that many covariates with this data. You might at best chase a couple, but I'm doubtful about even that. All the missing observations mean that you have rather little information about detection probability (as evidenced by the SE on p in the model results Bret posted for you) and it will be difficult to impossible to disentangle detection from the effects of covariates.

Also, the encounter histories you sent first do no match the input file you just sent. Lots of '.' are replaced with zeros... which are they really? There is a huge difference between a '.' (didn't do the survey = no information whatsoever) and a zero (went and did not detect = information about detection).

It seems like you need to do some more reading about occupancy model basics. The 2006 book by MacKenzie et al. would be a good start... the preparatory chapters at the beginning are excellent.

- Dave

[David]

yes, there was a problem with the first data i entered. the second encounter histories is the right one. and the p and psi values (and SE) are different:

Real Function Parameters of {p(.)psi(.)}
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p 0.6092218 0.0888994 0.4285295 0.7642166
2:Psi 0.4566654 0.0893676 0.2932359 0.6299892

to my knowledge the P SE is not that bad.
any way, what can be the reason that the SE of P and psi of some of the models equals zero?

thanks,

[bacollier]

David,
There are a variety of reasons why you could get nonsensical results from CMR data (as Dave pointed out). From your example above, your constant and constant model you show below gives what I would call an 'ok' SE on . But, when you run a model with 8 additive effects (9 total parameters) for 40 data points, you are overparameterizing the model quite a bit, probably part of the nonsensical answers you are given. Are all those covariates supposed to be affecting , or are some supposed to affect . From your below, it looks like you have them all affecting , correct? Also, you have variables that range from the 1000s to <10, so you might want to standardize your covariates, but I don't think that would help too much. As David suggested, have a look at Darryl's 2006 book, as well as the first few chapters in the MARKBOOK as it seems you need to better define your models in general. Also, do some searching on the phidot board on parameter estimability as that might help you out.

Bret