Hi Jhoana.

In first place, you must search a unimodal distribution from your data. In Bioesta 5,3 on statistic section, go to transformation and sortie and to use, Box e Cox, you obtained what is the better transformation function for your data base .

Regards

Dear Johana

It is probably better to rethink you sampling protocol, because with such a kind of distribution and the number of available detections it would be rather difficult to obtain reliable estimates. I think that some of the assumptions of distance sampling are not met in your case (I guess something related to the transect distribution and/or to the behaviour or you animals), but you have to check. I suggest to refer to the references listed in the distance web site: http://distancesampling.org/whatisds.html

Good luck

DG

Hi Johana,

You say “we have an increase number of detections in a larger distance” this seems a serious issue, but is difficult to understand what might be going on with the information you provide. What are you sampling and how? Animals? If you are sampling animals it might be the case that they are detecting the observer before the observer sees them, and running away. Similarly, your objects might be reacting to whatever you use as transect line.

Alternatively, if you arrange the data in strips of different width (the further the larger), you might expect to get more observations there, despite lower detectability, because the surveyed area is larger.

I am sorry, but we cannot provide better advice without more information.

Regards,

Jabi

Thanks you all for your help. Additional information:

I am sampling monkeys, the specie Callicebus (plecturocebus) caquetensis, in colombia, south america. We set five transect, three of them through the largest fragment, which is 120 ha. The other two transect, was stablished along the riparian forest on the river and the second on a higher altitude that comprise terra firme forest and inundated in the lowest part of the transect. In general, all of them have the inundated forest condition... except the one on terra firme. Furthermore, each transect had a w: 75 m and a 2w:150 m; the last was the distance among transects. The inundated condition made it difficult for me to extend my sample size.

I can only provide the histogram that distance generate with the key funtions. Sayan by the original histogram you mean the frequency plot of data?

Thaks again.

Best regards.  

Well, it might be that monkeys are detecting you and moving away. It might also be that if you are using some kind of structures to perform the transects (roads, paths, rivers) monkeys could actively avoid them or forest border areas and thus you are not detecting them in the closest strips.

This said you can feed Distance with different groupings to produced histograms (either by creating cut points within distance, or by re-arranging the data before imputing them into Distance). You could try creating a large strip in the vicinity of the central transect line, assuming that the “observer disturbance” is stronger in the central area, and that should correct, at least partially, for it. Depending what your aim is there might be other ways you could solve this. But, anyway, it is very important that you understand what is happening to your data and why are you observing these results.

Best,

Jabi  

Sayan and researchers here is the frequency histogram...axe 'x' perpendicular distances and axe 'y' frequency of observations at this distance.. as you all can see we dont have a good size sample. 

Hi Johana,

That’s not bad!

There is always some variation in the sampling by sampling error and process error. The distribution of the animals is not totally random in the landscape. Considering the small sample size you have, I would say the distribution is pretty good. You could try merging cols 2 and 3 or enlarging col 2 (make it 9 to 20 and col 3 20-25) and see what happens. However, I have seen much-much worse than that. Have a look to the statistics Distance produced for you and the goodness of fit of the detection function to your data. I would say it’s good enough.

As you can see it is not the histogram of the detection function, Jabi. It would be the best plot if it were so...

I was wrong when i said that at a larger distance more observations. I tried to transform data via box cox tool, it made data fit in the normal but when i plotted the histogram it turns out to be totally different. Before using this tool it does also fit in normal with a P:0.0532... Almost out of normality. So, will it be better to transform data with Box Cox suggestion or just leave data as it was taken on field?.

I hope you all can understand me.  

Hi Johana,

I do not really think that it is a serious problem. Data often depart form normality when they are scarce. In mi opinion “your problem “ is that you have little data and columns, therefore if two of them depart from expected the fit will be poor. Think of throwing a coin: there are two possible outcomes and each has 0.5 p. Well if you throw it just once you will not have 0.5 of each. If you throw twice you might or not, but a 0.5-0.5 distribution will only happen with hundreds, thousands of trials. In your case, you can say that for the sample size you have, you cannot say that it departs from normality, and move on.

Alternatively, try asking Distance to make more than 5 bins (6,7,8,9) and see if you get a better distribution. You can arrange the bins manually to try to adapt to particular problematic data (i.e your 9-15 interval might be low because there are a few individuals in 8 or 16, try making it larger 9-16 and see if that corrects for it. Finally, in addition to that, I guess 42.201 is the distance to your furthest observation. You can try to truncate at 42 or to allow larger distances (ie. Starts:0, ends:50 to try to model better the tail of the distribution) .

You need to play with your data,

Cheers,

Jabi

I wouldn't transform the data, but looking at your histogram so suggests to me your data aren't too bad. If they were you could try left truncation (problematic), but as it is I wouldn't bother.  With less than the recommended 60 observations, this kind of distribution occurs reasonably frequently.

You may be interested

Article PERCENTILE BASED HISTOGRAM BIN WIDTH