I have an amalgamation of independently collected datasets covering a large area of ocean where surveys to count birds have occurred sporadically over a long time. Some surveys were from boat, some were from air, and often there is some spatial overlap between these efforts.

I want to link the bird counts to environmental covariates, and use that relationship to make predictions about abundance in areas that have not been sampled based on their environmental conditions. To control for variability in counts due to non-environmental data, I also wanted to include additional information such as vessel type (plane vs boat) but, given how the environmental data have been collected--there is a grid for the entire study area, each grid cell has one row of environmental data--I am not sure that this is a statistically sound thing to do. Environmental variables are all static (e.g., distance to shore, depth to seafloor, etc) so they are not expected to change through time. I have also summarized the total count and effort for each surveyed grid cell such that it can be linked to the gridded environmental data. When I think about splitting these rows based on grid cell further by vessel type, I can't help but wonder if this is a form of pseudoreplication because the environmental covariates for a given grid cell would be identical.

I am familiar with examples of pseudoreplication, like "treating multiple leaves from the same plant as replicates; treating multiple plants from the same pot or flat as replicates" but I am not sure if what I have here would also be considered pseudoreplication -- would this be equivalent to treating multiple counts from the same location as replicates...?

Here is a hypothetical dataset to better illustrate my question:

  • Database that excludes vessel type as a covariate:
Grid ID Count Distance to shore (m) Distance to seafloor (m)
1 25 98.7 204.6
2 2 1049.2 1027.8
  • Database that includes vessel type as a covariate:
Grid ID Count Vessel Distance to shore (m) Distance to seafloor (m)
1 20 Boat 98.7 204.6
1 5 Plane 98.7 204.6
2 1 Boat 1049.2 1027.8
2 1 Plane 1049.2 1027.8

Any advice would be helpful!

  • $\begingroup$ To me, this does not sound like pseudoreplication $\endgroup$ Commented May 19, 2022 at 18:00
  • $\begingroup$ Please edit the question to say more about what you are trying to predict and what variables you are using beyond those you show. This would seem to require including time in some form as a predictor, for example. Depending on the model, you might be able simply to use Vessel as a predictor variable. $\endgroup$
    – EdM
    Commented May 19, 2022 at 18:38
  • $\begingroup$ @EdM I attempted to address your request by editing this question. Maybe I'm overthinking it and I can just have multiple rows for each grid cell even though the environmental data are duplicated each time there is >1 row for a given grid cell. There is no time component in the model, all the data are being treated as if they were from the same time because there isn't interest in understanding interannual variability or trends. There is a spatial component, specifically latitude and longitude for each grid cell centroid. Most of the other variables are related to distance. $\endgroup$
    – Wu Wei
    Commented May 19, 2022 at 20:41

1 Answer 1


To put this in the context of Hurlbert's classic paper on "Pseudoreplication and the Design of Ecological Field Experiments," Ecological Monographs 54: 186-211 (1984), you have what he called a "mensurative experiment":

Mensurative experiments involve only the making of measurements at one or more points in space or time; space or time is the only "experimental" variable or "treatment."

In mensurative experiments generally, pseudoreplication is often a consequence of the actual physical space over which samples are taken or measurements made being smaller or more restricted than the inference space implicit in the hypothesis being tested.

That might be what led you to worry about pseudoreplication, as the repeated measurements within a single grid cell are much more restricted in area than the overall grid.

Provided that you are handling spatial autocorrelation in your grid correctly, however, then you shouldn't have to worry about the overlap in observation areas. Zuckerberg et al., in "A Review of Overlapping Landscapes: Pseudoreplication or a Red Herring in Landscape Ecology?" Curr Landscape Ecol Rep 5, 140–148 (2020), put it pretty clearly:

We suggest that the perceived problem of overlapping landscapes is distinct from more important issues in landscape ecology, such as a robust sampling design complete with a discrete assessment of spatial autocorrelation. Through simulation, we demonstrate that changing the amount of landscape overlap does not alter the degree of spatial autocorrelation.

If you include Vessel as a predictor then this is even less of a problem, as you will be modeling differences between independent Boat and Plane observations explicitly--for which such overlap areas might be particularly informative.

I do worry about completely ignoring time in your model. It seems that at least some seasonal analysis might be called for. If your understanding of the subject matter indicates that seasonal differences in bird counts aren't important in this region of ocean, OK, but be prepared to defend that position strongly.

  • $\begingroup$ Thank you for the in-depth response and citations @EdM! Actually there is a seasonal component as well, there will be different models for different season since the way birds relate to their environment differ by season. I just wanted to provide as simple of a scenario as I could to highlight the specific question I was struggling with. $\endgroup$
    – Wu Wei
    Commented May 23, 2022 at 16:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.