Northern Prairie Wildlife Research Center
Mallard Recruitment in the Agricultural Environment of North Dakota
Evaluation of Assumptions
Radios Did Not Influence Breeding Behavior
Gilmer et al. (1974) reviewed the difficulties in determining the effects of
radios on free-flying wild ducks, but their results suggested that a radio transmitter
does not substantially affect reproductive effort during the spring and summer.
We do not believe that transmitters had an effect on nest site selection or
nest success because the data from our studies are comparable to results obtained
with other techniques in the same area. It is more difficult to evaluate the
impact of the transmitter on hen success because there are few other techniques
available for measuring hen success. If the radios do have an effect, they probably
influence the probability of hens initiating either the first or subsequent
nests. We used the relation H = fPef(1
- P)2 from the Cowardin-Johnson (1979) model
to compare hen success from the model with the observed hen success. In the
wet years 1978 and 1979, when f should be near 1, hen success was 23.7%
and 17.4%; the model predicted 27.8% and 16.0%, respectively. If the underlying
assumptions of the model are correct, this result suggests that the radios did
not greatly affect nesting effort. No conclusive statement can be made about
the effect of radios on our recruitment estimates, but there is no evidence
of major impact. If an effect does exist we would expect the bias to cause us
to underestimate recruitment.
The Birds Captured Are Representative of the Population
Two factors that may bias results obtained from a sample of marked birds are
selection of birds for trapping and susceptibility of birds to the trapping
methods. It was not possible to obtain a truly random sample of birds. We attempted
to trap birds throughout the study area but trapped only 3 hens in the Drift
Plain stratum because water and waterfowl populations were sparse. Our sample
is, therefore, much more representative of the Coteau than of the Drift Plain.
We also frequently trapped birds near roads because the ease of travel made
it possible for us to operate additional trap sites and obtain more birds for
marking. Subsequent observation of marked birds demonstrated that these birds
moved about on a number of ponds with no apparent preference for roadside ponds.
Furthermore, there was no indication that birds trapped near roads nested in
roadside habitat more frequently than birds trapped away from roads.
A trap bias can exist in a sample when 1 age class of birds is more susceptible
to trapping than another. The only estimate we have for age ratio in the population
is based on our trapped sample; therefore the bias can not be evaluated. We
were able to compare the age ratio in the samples obtained with rocket nets
with those obtained with the shoulder-held, net-firing gun. The gun sample
had a significantly higher proportion of SY birds than the rocket-netted sample,
perhaps because the SY hens were less wary than the ASY hens. Possibly, SY
birds also were more susceptible to capture by rocket net than ASY birds,
but we have no means of testing this hypothesis. If such an age bias existed,
it would cause us to underestimate recruitment because we have demonstrated
that older hens are more successful than yearlings.
Birds in the Population Did Not Breed Elsewhere
If a portion of the birds selected for our breeding population bred elsewhere,
our recruitment estimates would underestimate the contribution to the continental
population although they would still be meaningful for the local study area.
We have no evidence that birds included in our sample made extensive moves to
areas where their breeding effort would go undetected. On the contrary, our
telemetry data show that the birds were faithful to the local area where they
were captured as long as the hen remained reproductively active. When breeding
activity ended after the destruction of 1 or more nests, the hens joined flocks
of mixed sexes. These nonbreeding birds tended to drift across the study area
and frequently left the area completely. Gilmer et al. (1977) demonstrated some
northward movement by postbreeding females. One of 18 direct recoveries of our
hen mallards was from Alberta.
No Successful Nests Went Undetected
This is probably the most important assumption underlying our recruitment estimates.
Table 14 shows that at the low observed hen
success, failure to find a hatched nest would result in a serious negative bias.
We know of no method that enables the investigator to follow the fate of all
nests for a group of hens and thus to evaluate both the sampling methods and
the many biases that may influence results. To overcome this problem we used
results of a simulation model. The model is stochastic and employs probability
functions based on field data according to the flow diagram illustrated in Figure
11. It covers the period from the time the birds arrive until the young
are fledged and accounts for variation in weather, age structure of the population,
and predation on eggs and hens. We entered values for nest survival and habitat
selection probabilities derived from our data. The results from the simulation
agreed well with the input parameters. This gave us confidence that the model
could be used to test for biases in sampling procedures. Among the outputs of
the model is a chart showing the nesting history for each hen. The advantage
of the simulation is that it furnishes a known population from which to sample.
We know of no way to obtain complete histories of all nests in a real population.
A small portion (10 hens selected to demonstrate the method) of the output
for 400 simulated hens that initiated 819 nests is shown in Figure 12. We
have assumed that the model represented a population of nests similar to the
real population that we sampled and then simulated marking birds and obtaining
radiolocations in a manner similar to the methods used in the field. We first
determined probability of capture on a given date from our distribution of
captures and thus determined a capture date for each hen entering the simulated
sample. Next we simulated 14 tracking flights over the population at 6-day
intervals. We assumed that we would find the marked bird and that the nest
would be found if the hen was on the nest. Then using data on nest attentiveness
presented by Caldwell and Cornwell (1975) and our data on the time of day
when flights were conducted, we postulated the probability of finding a nest
as a function of age of the nest. The results indicate that the probability
of finding a nest is very small early in laying and high late in incubation.
The probability was used to determine if nests were found when the hen was
found. The estimates from our simulated sample from the simulated population
are similar (Table 19).
Table 19. Comparisons of simulated sample data and simulated population
of mallard nests.
||Sample from simulated
||True value simulated
|Percent of total
||Percent of total
Of the 819 nests in the simulated population, 378 were found by the simulated
radio tracking. Eighty-one of the nests in the simulated population were successful
and all of these were found by the simulated sampling procedure. Successful
nests are found because once the hen enters incubation she is usually on the
nest and, because searches were at 6-day intervals (about the same as our
field conditions), there are at least 4 chances to find a nest that is incubated
for the full 26 days. Note that successful nests of hen numbers 1 and 7 (Fig.
12) were found on the second and third searches after incubation had begun.
In summary, there are potential biases inherent in the field methods used
in this study as there are in any study that requires capture and marking.
These biases cannot be estimated because the population parameters of interest
require capture and marking for estimation. The fact that some estimates derived
in this study lead to the same conclusions as data gathered by other studies
strengthens our confidence in the data. A review of the underlying assumptions
of our study and comparison of our results with results based on a simulation
model suggest that our results are logical and consistent. These analyses
do suggest that the potential biases in this study would cause us to underestimate
rather than overestimate recruitment.
Previous Section -- Models and Tests of Assumptions
Return to Contents
Next Section -- Biases in Analysis Techniques