Hello Rehan

Before I hazard an answer, I was wondering if perhaps you could clarify a few points for me? You are working within a single lake at 3 different points. Each of the different points is a different habitat, so your sample size for each habitat is n = 1. Are you attempting to say anything about habitat type? If you are, then I am not sure if n = 1 is adequate. If possible (and sometimes its not) you may want to increase that.

At your three different points, are they different colonies, so in effect you are looking at 3 colonies?

What behavioral data are you collecting? Proportion time spent doing "X", occurrence of "X", etc?

What is your sample size? More specifically, how many observations do you have at each point? Keep in mind that your ability to speak to habitat type is going to be limited, as you only have a single replicate in each habitat type. However, even though you only have 1 replicate per habitat type, if you have ample replication to create that 1 data point, then at the very least it is a high quality data point.

All the best

Travis

Hi Rehan,

Good question. Travis raises great questions too. Yes, I also interpreted your description of the data structure as the following:

1 lake / 3 habitats per lake / 1 colony per habitat

If there's one colony in each habitat, then you will have less power to generalize the patterns you find to outside the lake environment. That is, you will not have variation due to multiple replicates of each habitat (as Travis asked about). Perhaps you could clarify your design a bit more?

Anyway, if you are looking at proportion time spent in each activity, my perspective is that scan sampling (as outlined in Altmann's 1974 paper on quantifying behavior) has some major drawbacks. If you quantify presence/absence of behaviors X,Y, and Z every five minutes (scan sampling), then you may miss some behaviors that are in your ethogram. Scan sampling can be sensitive to phase-matching between scan sampling and behavioral cycles (if they are cyclical), especially for behaviors shorter than the scan sampling interval. For instance, if a behavior occurs for one second every five minutes and the scan sampling is matched to this interval, then the proportion of time spent in that behavior will be measured as identical to the proportion of time spent in a constantly-occurring behavior (a 99.2% error). Though this is an extreme case, it does point out that other sampling methods such as all-occurrence sampling may be best. An advantage of all-occurrence sampling may allow you to capture more within-colony/within-habitat variation. Simply pick individuals out of the colony and then record all behaviors for 15-30 seconds. Then, repeat the process sequentially on different individuals. Again, this may not be appropriate for your data or what's relevant or feasible for your study population, but these are some additional perspectives.

Another thought, perhaps use a video recorder to capture images of the behavior. You can then collect a large amount of data and go over it at a later date. I have never done it myself, but members of my lab have. Check out their publication for more details:

MacDonald, E. C., et al. (2012). "Variability in Foraging Behavior and Implications for Diet Breadth among Semipalmated Sandpipers Staging in the Upper Bay of Fundy." The Condor 114(1): 135-144.

I like Travis' idea for the use of a video camera. Use of one video camera per habitat would also allow you to collect data simultaneously. This would allow you to examine differences in behavior throughout the day or in response to rare events (a predator flying over, etc.). For studies of individual behavior, I usually avoid using videos, as this just adds a bunch of work after the fact (and prevents one from paying attention in the field), but I think Travis is right that it may be the only way to capture the data from a bunch of individuals within the colony.

On a related note, I'm just really glad no one has responded in the typical fashion by citing the issue of non-replication/pseudoreplication/etc. Sometimes, nature doesn't give you the kind of replication that you might want, but it doesn't mean that good inference can't be drawn about patterns and possible processes within such systems. Anyway, it sounds like a cool project. Keep us informed.

Thank you so much Travis and Patrick for your replies and guidance,

Exactly Patrick you are right it is 1 lake / 3 habitats per lake / 1 colony per habitat

Let me clear all the questions

Actually due to the large size of Lake (300 km²) and due to great diversity of water depth Lake has different vegetation zones, I divide the lake into three zones keeping in view different vegetation zones. (Emergent plants, Floating plants, Marginal plants,scrubs etc) So i am trying to cover all those vegetation zones and behavior of waterbirds in those zones...And from waterbirds i mean all species of waterbirds present and engaging in any behavior at time during observation.

So we can also say that Lake is my study area and we further divided into 3 habitat types and we are studying behavior of all species present at given time in each habitat.

In collecting behavioral data i am following "Behavior of migrant shorebirds in playas of the Southern High Plains, Texas"CA Davis, LM Smith - Condor, 1998" but there are some changes also as we have different species here.

We take 1 observation at one point per month(3 observations/ month) , at each point we take observation from dawn to dusk, it means total app 144 observations per day.. I have divided daylight hours into 4 time intervals: dawn (sunrise-0800), morning (0800-1200), afternoon (1200-1600), and evening (1600-sunset).

About Video Cameras...It was my 1st thought also but due to the heavy fishermen activities in the lake cameras are not safe to leave and distance i as i use binoculars or field scope to observe waterbirds (from that distance camera cannot work ) and even then if i install camera in near the waterbirds then i have to install a lot of cameras to observe all birds) so its not feasible.but i wish i could as sitting all day alone with binoculars is quite a tough job :p

Hope it clears all the questions :)

Regards

Hello Rehan

Wow, your project sounds really cool! It is also sounds like you have good replication! As for what kind of model you should use, your set up enables you to use many different methods. I have never created such models, so I would lean towards the literature for guidance. However, I think you could make some very interesting regression models. For guidance on making high quality regression models I suggest the following book:

Burnham, K. P. and D. R. Anderson (2002). Model selection and multimodel inference: a practical information-theoretic approach. New York, New York, Springer Verlag.

Best of luck

Travis

Hi Rehan-

Given that you are looking at a disparate components of a single system, you might think about a mixed-effects regression format, where the 3 areas represent the random effect and then whatever variety of biotic parameters you are hypothesizing about would make up the fixed effects. This would at least let you come up with some estimates for the whole system with proportionate inference from the differences between sites.

I agree with all of the posts above as well - multi-model inference is a great way to go to sort through this potentially wide array of data and inputs.

We did a paper a few years ago with this same kind of mixed-effects regression and multi-model competition that might give you some idea of potential model structure:

Ransom, J.I., Cade, B.C., and N.T. Hobbs. 2010. Influences of immunocontraception on time budgets, social behavior, and body condition in female feral horses. Applied Animal Behaviour Science. 124: 51–60.

You can find full text here: http://www.fort.usgs.gov/Products/Publications/22643/22643.pdf

Hope that helps!

best of luck-

Jason

Thankyou so much Travis and Jason...I will read the provided literature.

Thanks again :)

Absolutely, a mixed effects approach is the only way to go in order to account for the clustering in the data. It seems you will also be taking data at specific bouts (e.g. 1 hour in morning and 1 hour of afternoon and maybe across multiple days), so you may have a more complex hierarchical structure (bout nested within site)

Depending on the temporal scale your measurements, you may also consider an autoregressive structure to your data.

” I want to see the difference in behavior among three points”

OK, but what is the working hypothesis?

P.S. I mean: the (assumed) reason for differences in behavior... due to... what?

I am really sorry i didn't explain this point ....

Due to difference in Habitat structure (due to different vegetation) , Months (as data will be collected for 12 months), daily time (as i divided the time into 4 quarters) ,Temperature and Water level.

Patrick actually its a full day observation with day hours divided into 4 intervals dawn (sunrise-0800), morning (0800-1200), afternoon (1200-1600), and evening (1600-sunset).