path analysis?

[bowlerbear]

Corey and I discussed whether a path-analysis style analysis might help model/visualize the links between land-use, species richness, number of checklists and sampling coverage. I had a quick and dirty go at this using data for 2019/20 (and just for q=0).

(note: I didnt orginally include urbancoverfraction in the model for number of checklists but the psem suggested strong support for this link, so I added it post-hoc).

library(piecewiseSEM)
psem1 = psem(
  lm(sampling_coverage ~log(number_checklists), data=output0),
  lm(log(number_checklists) ~ qD + urbancoverfraction, data=output0),
  lm(qD ~ heterogeneity + urbancoverfraction + waterpermanentcoverfraction +
       treecoverfraction,data=output0)
)

summary(psem1, .progressBar = FALSE)
plot(psem1)

But number of checklists here is just the observed number of checklists. Not the number of checklists needed to sample to a certain level.

[sablowes]

Agreed that something along these lines could be useful for conceptual illustration, but this is a bit hard to parse and provide constructive feedback on without more context, sorry.

• the arrow between number_checklists and sampling_coverage suggests that sampling effort (i.e., number of checklists) has not been standardised prior to calculating coverage. My recollection from the last discussion with Corey, Diana and myself was that this was necessary for the goal of predicting how many and where future checklists should be collected to describe diversity?

• But maybe you are trying to do/show something else with this? e.g., if this is a conceptual overview of relationships among environment, diversity, sampling effort and sample completeness, then it is looking good! My only comment would be that qD should be in the path to sampling_coverage. And we would probably also not expect diversity and sampling effort to have independent effects on estimates of completeness: low sampling effort and low diversity would often still result in relatively 'complete' samples, but low effort and high diversity will most often lead to low completeness. But I've no idea whether interactions even go in path models! So perhaps it is sufficient just to include diversity in the path to completeness, and not worry about any interaction.

Two cents spent ;)

[bowlerbear]

yeah, you're right. I think I have morphed Corey's original question into my own question about predictors of sampling effort :) Its interesting to me that most of the effects of land-use on sampling effort can be explained indirectly, by diversity, except urban cover.

Anyhow, we have gone down multiple analysis avenues and still seem to be a but stuck. Its mostly about whether we just throw all the variables shown in the path diagram together in one big model. Or separate them in some logical way like above - for ease of understanding. There is also some question about direction of arrows too.

Its probably best if we get all our results together and show you them to help us decide which method. And stop us procrastinating any more.