How is censoring individuals actually working?

[kernicholson]

I was given advise that it would be reasonable to use a CJS platform to obtain survival estimates as long as my resight parameter = 1. But I have the opportunity to increase the sample size if I use KF platform by including censored individuals. i.e., those that had transmitter failure. With the CJS model - I simply removed those individuals completely from the data set. Therefore, sample size would be 196 for KF and 126 for CJS.

I ran both KF and CJS on the exact same 126 samples I get the same results/models/survival estimates etc.

When I calculate survival using 126 individuals (no censored animals)
Interval 1 (summer) survival is 0.959
Interval 2 (Winter) survival is 0.951

When I calculate survival using 196 individuals (including censored animals) using KF

Interval 1 (summer) survival is 0.93
Interval 2 (winter) survival is 0.96

for KF with 3 encounters, 2 intervals or "seasons"the possible capture histories are:
1010 = survived the entire study (both intervals, summer and winter)
1100 = died by the 2nd survey (died in summer interval)
1011 = died by the 3rd survey (survived summer, died in winter interval)
1000 = survived summer, but censored in second interval (winter) those who were lost because of transmitter failure or they were removed from the study for some other reason.


Why would summer survival decrease and winter survival increase? Or is it more of the summer survival is decreasing and winter is staying approximately the same? Why is summer survival decreasing when all that happened was an increase of individuals to the sample size that have all survived the first interval (coded as 10 in the 1000 capture history).

Say I had 100 animals, 60 of them had 10 code to survive interval 1, 40 had code 11 for that same interval.
Now increase the sample size to 150, now 110 have a survival coded 10 for interval 1 and still 40 are coded as 11. There were no new deaths in the interval, so wouldn't the survival estimate increase?