2.3. Data analysis
To obtain unbiased estimates of mean IBI, both surface dives and deep dives need to be recorded with the same probability.
If the data sampling is more likely to record one of the two dive types, this can lead to the mean IBI estimates either being positively (if more deep dives are recorded) or negatively (if more surface dives are recorded) biased.
While surface dives within a bout are relatively easy to record from the same individual, the long duration of deep dives often makes it hard to predict where and when the whale will surface after the dive.
When several whales are present in the same area (off Sri Lanka at times up to ten individuals) the chances of confusing the focal whale with another whale, and thus missing the surfacing time, is also relatively high.
As a result, many focal follows ended up constituting only a single dive cycle, and some just a single surfacing bout. To overcome this sampling bias when estimating mean IBI of the follow, the temporal dependence between dive types within follows was estimated, using a first-order Markov chain.
The time series data of dive types, one for each follow, were first compiled into two-way contingency tables of preceding dive type versus succeeding dive type using R.
If a follow ended with a high arch or a fluke (indicating the transition from a surfacing dive to a deep dive), an additional transition from surfacing dives to deep dives was added to the contingency table to reduce any sampling bias from missing the longer deep dives.
For the same reason, a transition from deep dives to surfacing dives was also added, since a deep divewas never observed directly following another deep dive in this dataset.
Transition probabilities from preceding to succeeding dive types were then calculated using the following equation
where i is the preceding dive type, j is the succeeding dive type, n is the total number of dive types (i.e. two), a
is the number of transitions observed from dive types i to Journal of Experimental Marine Biology and Ecology
is the transition probability from i to j in the Markov chain.
To test whether or not the estimated contingency table differed froma theoretical distribution, a goodness of fittest was performed using Pearson's chi-squared test in R.
2.3. Data analysis
To obtain unbiased estimates of mean IBI, both surface dives and deep dives need to be recorded with the same probability.
If the data sampling is more likely to record one of the two dive types, this can lead to the mean IBI estimates either being positively (if more deep dives are recorded) or negatively (if more surface dives are recorded) biased.
While surface dives within a bout are relatively easy to record from the same individual, the long duration of deep dives often makes it hard to predict where and when the whale will surface after the dive.
When several whales are present in the same area (off Sri Lanka at times up to ten individuals) the chances of confusing the focal whale with another whale, and thus missing the surfacing time, is also relatively high.
As a result, many focal follows ended up constituting only a single dive cycle, and some just a single surfacing bout. To overcome this sampling bias when estimating mean IBI of the follow, the temporal dependence between dive types within follows was estimated, using a first-order Markov chain.
The time series data of dive types, one for each follow, were first compiled into two-way contingency tables of preceding dive type versus succeeding dive type using R.
If a follow ended with a high arch or a fluke (indicating the transition from a surfacing dive to a deep dive), an additional transition from surfacing dives to deep dives was added to the contingency table to reduce any sampling bias from missing the longer deep dives.
For the same reason, a transition from deep dives to surfacing dives was also added, since a deep divewas never observed directly following another deep dive in this dataset.
Transition probabilities from preceding to succeeding dive types were then calculated using the following equation
where i is the preceding dive type, j is the succeeding dive type, n is the total number of dive types (i.e. two), a
is the number of transitions observed from dive types i to Journal of Experimental Marine Biology and Ecology
is the transition probability from i to j in the Markov chain.
To test whether or not the estimated contingency table differed froma theoretical distribution, a goodness of fittest was performed using Pearson's chi-squared test in R.
การแปล กรุณารอสักครู่..
