negative values for the lower 95% limit of estimated psi

[alqamy]

I am modeling occupancy for antelope species over 30 sites and using 4 site covariables. The estimated lower level of the 95% confidence for individual psi values are negative values and greater than unity (1< without the sign).

Individual Site estimates of Psi:

Site Survey Psi Std.err 95% conf. interval
1 Arb Frst 1 1-1: 0.9965 0.0214 -0.4653 - 0.9786
2 Aslab St 1 1-1: 0.2607 0.1226 -8.7175 --0.1284
3 QS FRST 1 1-1: 0.9505 0.1760 -2.3758 - 0.9672
4 Release site 1 1 1-1: 0.5991 0.1847 -1.9377 - 0.5926
5 Release site 2 1 1-1: 0.7317 0.1393 -0.6476 - 0.6494
6 Release site 3 1 1-1: 0.6551 0.1843 -1.5121 - 0.6488
7 Shade 28 1 1-1: 0.2801 0.1702 -11.4703 - 0.2573
8 Shade 29 1 1-1: 0.2596 0.1699 -13.8543 - 0.2550
9 Shade 30 1 1-1: 0.7305 0.1390 -0.6488 - 0.6472
10 Shade 31 1 1-1: 0.1994 0.1189 -15.6385 --0.2150
11 Shade 32 1 1-1: 0.3886 0.1383 -4.0041 - 0.1744
12 Shade 33 1 1-1: 0.5180 0.1823 -2.7845 - 0.5039
13 Shade 34 1 1-1: 0.5553 0.1646 -2.0214 - 0.4960
14 Shade 35 1 1-1: 0.6726 0.1864 -1.4335 - 0.6703
15 Shade 36 1 1-1: 0.5135 0.1466 -2.1501 - 0.3887
16 Shade 37 1 1-1: 0.2784 0.1621 -10.8163 - 0.2048
17 Shade 38 1 1-1: 0.3140 0.1512 -7.3473 - 0.1568
18 Shade 39 1 1-1: 0.3196 0.1245 -5.5920 --0.0230
19 Shade 40 1 1-1: 0.3523 0.1046 -3.7730 --0.1105
20 Shade 41 1 1-1: 0.3515 0.1157 -4.1748 --0.0299
21 Shade 42 1 1-1: 0.3515 0.1160 -4.1869 --0.0280
22 Shade 43 1 1-1: 0.4960 0.1120 -1.7559 - 0.2270
23 Shade 44 1 1-1: 0.2670 0.1429 -10.0144 - 0.0587
24 Shade 45 1 1-1: 0.2360 0.1329 -12.1847 --0.0470
25 Shade 46 1 1-1: 0.1828 0.1078 -16.8211 --0.3763
26 Shade 47 1 1-1: 0.1828 0.1156 -18.5780 --0.2540
27 Shade 48 1 1-1: 0.6343 0.1936 -1.7885 - 0.6455
28 Shade 49 1 1-1: 0.0523 0.0595 -182.6934 --0.8833
29 Shade 50 1 1-1: 0.2293 0.1180 -11.1542 --0.2126
30 Shade 51 1 1-1: 0.6632 0.2204 -1.9864 - 0.7173
I wonder what does that mean. Any help is appreciated. Regards

[darryl]

When did you last update PRESENCE? The latest version has calculated confidence intervals on the logit-scale for some time now to keep the limits within the bounds of 0-1.

[alqamy]

I downloaded PRESENCE 2.4 and did the same run. Numbers are more or less OK with the upper confidence limit but the problem is still there for the lower limit. Is this something that I can overlook as long as the psi estimate is OK and this is due to large SE? Thanks in advance


Individual Site estimates of Psi:

Site Survey Psi Std.err 95% conf. interval
1 Arb Frst 1 1-1: 0.9965 0.0214 -0.4653 - 0.9786
2 Aslab St 1 1-1: 0.2607 0.1226 -8.7175 --0.1284
3 QS FRST 1 1-1: 0.9505 0.1760 -2.3758 - 0.9672
4 Release site 1 1 1-1: 0.5991 0.1847 -1.9377 - 0.5926
5 Release site 2 1 1-1: 0.7317 0.1393 -0.6476 - 0.6494
6 Release site 3 1 1-1: 0.6551 0.1843 -1.5121 - 0.6488
7 Shade 28 1 1-1: 0.2801 0.1702 -11.4703 - 0.2573
8 Shade 29 1 1-1: 0.2596 0.1699 -13.8543 - 0.2550
9 Shade 30 1 1-1: 0.7305 0.1390 -0.6488 - 0.6472
10 Shade 31 1 1-1: 0.1994 0.1189 -15.6385 --0.2150
11 Shade 32 1 1-1: 0.3886 0.1383 -4.0041 - 0.1744
12 Shade 33 1 1-1: 0.5180 0.1823 -2.7845 - 0.5039
13 Shade 34 1 1-1: 0.5553 0.1646 -2.0214 - 0.4960
14 Shade 35 1 1-1: 0.6726 0.1864 -1.4335 - 0.6703
15 Shade 36 1 1-1: 0.5135 0.1466 -2.1501 - 0.3887
16 Shade 37 1 1-1: 0.2784 0.1621 -10.8163 - 0.2048
17 Shade 38 1 1-1: 0.3140 0.1512 -7.3473 - 0.1568
18 Shade 39 1 1-1: 0.3196 0.1245 -5.5920 --0.0230
19 Shade 40 1 1-1: 0.3523 0.1046 -3.7730 --0.1105
20 Shade 41 1 1-1: 0.3515 0.1157 -4.1748 --0.0299
21 Shade 42 1 1-1: 0.3515 0.1160 -4.1869 --0.0280
22 Shade 43 1 1-1: 0.4960 0.1120 -1.7559 - 0.2270
23 Shade 44 1 1-1: 0.2670 0.1429 -10.0144 - 0.0587
24 Shade 45 1 1-1: 0.2360 0.1329 -12.1847 --0.0470
25 Shade 46 1 1-1: 0.1828 0.1078 -16.8211 --0.3763
26 Shade 47 1 1-1: 0.1828 0.1156 -18.5780 --0.2540
27 Shade 48 1 1-1: 0.6343 0.1936 -1.7885 - 0.6455
28 Shade 49 1 1-1: 0.0523 0.0595 -182.6934 --0.8833
29 Shade 50 1 1-1: 0.2293 0.1180 -11.1542 --0.2126
30 Shade 51 1 1-1: 0.6632 0.2204 -1.9864 - 0.7173

[jhines]

Did you get a warning/error message before the beta estimates about the variance-covariance matrix in the output? Even if you didn't, I would not trust estimates when the SE's are unreasonable. I suspect the model is overparameterized. Could you post the beta estimates and SE's?

Jim

[darryl]

Also, which model are you actually trying to fit to the data?