cwwhitney / ethnobotanyr Goto Github PK

We discussed the incorporation of spatial analysis. Maybe a tall order but we should at least think about it.

The papers:

Harou et al. - 2020 - Mapping Flood-Based Farming Systems with Bayesian .pdf
Liman Harou et al. - 2021 - Crop modelling in data-poor environments – A knowl.pdf

Start with Eight to Late
https://eight2late.wordpress.com/2015/05/27/a-gentle-introduction-to-text-mining-using-r/

FLs is calculated correctly, as in the cited paper. But it is stated wrongly in the vignette.
Where it states URs it should state FCs instead, as this is what you actually do.

It should also state that the values are multiplied by 100, both in the vignette and in the help file.

@CWWhitney, Thank you for the development of the package! Is it possible to change the colors? which function do you use to change the colors of the CIs-plot?

Originally posted by @KamilaBio in #21 (comment)

The calculation of both CIs UVs and simple UVs seems over excessive.
They all calculate the exact same thing correctly - but it seems error prone to do it in three different ways.

All three implementations calculate the exact same thing, and I would suggest using only one function and letting the other functions just refer to that one function.

The implementation of CIs looks cleanest and I would suggest the others just pointed to that.

Due to the use of this function:

RFCdata <- RFCdata %>% dplyr::mutate_if(is.numeric, ~1 * 
                                          (. != 0))

The function doesn't work if the informant column is numeric. I am afraid that this error might occur elsewhere as well, but I have not come across it so far.

The line can simply be replaced with, which would work:

RFCdata$FCps <- (RFCdata$FCps > 0) * 1

CVe gives the wrong result + it gives a warning because of the bind_cols command.

1. It binds columns together with rows in different orders.
2. The final calculation is where you multiply by a sum, I think is wrong - but a choice.
   a. You calculate it as stated verbatim in Reyes-Garcia et al. 2006, but I think they made an error when they wrote it. I doesn't make sense or add any information to multiply all the values by the same number.
   b. The way Tardio and Pardo-de-Santayana 2008 quote it is more sensical. Where they don't take the sum. My correction is based on a shift from Reyes-Garcia et al. 2006 verbatim to the interpretation of CV_e (or CV_s as they call it) in Tardio and Pardo-de-Santayana 2008.
   c. The latter interpretation which I suggest would also follow the interpretation in B. Hoffman, T. Gallaher, Importance indices in ethnobotany. Ethnobot. Res. Appl. 5, 201–218 (2007).

I have highlighted the problematic sections below and my suggestions for fixing them.

CVe <- FCs <- UR_UN_FC <- CVe <- URdata <- data_Ci <- data_URs <- URps <- sp_name <- informant <- NULL
URdata <- ethnobotanydata
URS <- URs(URdata)
NUS <- NUs(URdata)
NUS$Uce <- NUS$NUs/ncol(dplyr::select(URdata, -informant, -sp_name))
FCS <- FCs(URdata)
FCS$Ice <- FCS$FCs/length(unique(URdata$informant))

# ERROR 1:
# REMOVE
# UR_UN_FC <- dplyr::bind_cols(NUS, URS, FCS)
# REPLACE WITH THIS:
UR_UN_FC <- dplyr::left_join(NUS, URS) %>% dplyr::left_join(FCS)

UR_UN_FC$IUce <- UR_UN_FC$URs/length(unique(URdata$informant))

# ERROR 2 (Maybe???)
# REMOVE
# UR_UN_FC$CVe <- UR_UN_FC$Uce * UR_UN_FC$Ice * sum(UR_UN_FC$IUce)
# REPLACE WITH THIS:
UR_UN_FC$CVe <- UR_UN_FC$Uce * UR_UN_FC$Ice * UR_UN_FC$IUce

CVe <- dplyr::select(UR_UN_FC, sp_name, CVe) %>% dplyr::arrange(-CVe) %>% 
  dplyr::mutate(CVe = round(CVe, 3))
as.data.frame(CVe)

Add more functions

• informant consensus factor (Fic)
• UV
• FL
• others?

Quantitative methods in ethnobotany and ethnopharmacology: considering the overall flora--hypothesis testing for over- and underused plant families with the Bayesian approach.
https://www.ncbi.nlm.nih.gov/pubmed/21767623

Spiritual and ceremonial plants in North America: an assessment of Moerman's ethnobotanical database comparing Residual, Binomial, Bayesian and Imprecise Dirichlet Model (IDM) analysis.

https://www.ncbi.nlm.nih.gov/pubmed/23541935

All functions seem to return a result using as.data.frame(...).

The built in example dataset is also a data frame class.

Since the package is marketed as an intended expansion package working with tidyverse is should probably be updated to tibbles.

The code doesn't seem to calculate the Relative Importance Index as in the paper you cite.

I think the RFCs(max) part of the code is wrong.

Your code says:
RFCs <- RFCdata %>% dplyr::group_by(sp_name) %>% dplyr::summarize(RFCs = sum(FCps/(length(unique(informant))))) %>% dplyr::arrange(-RFCs)

Which gives FCs over total number of informants not over the max(FCs). There are 20 unique informants for all species in the test dataset, even though not all of them mention all species.

For the test dataset your code gives this:

> ethnobotanyR::RIs(ethnobotanydata)
  sp_name   RIs
1    sp_c 0.925
2    sp_a 0.812
3    sp_d 0.800
4    sp_b 0.738

while the RI should be:

> RFCsmax <- FCs(ethnobotanydata) %>% mutate(RFCsmax = FCs/max(FCs))
> RNUsmax <- NUs(ethnobotanydata) %>% mutate(RNUsmax = NUs/max(NUs))
> 
> RI <- left_join(RFCsmax, RNUsmax, by = "sp_name")
> RI %>% mutate(RIs = (RFCsmax + RNUsmax) / 2) %>% arrange(desc(RIs))
  sp_name FCs   RFCsmax NUs RNUsmax       RIs
1    sp_c  17 1.0000000   8   1.000 1.0000000
2    sp_a  15 0.8823529   7   0.875 0.8786765
3    sp_d  12 0.7058824   8   1.000 0.8529412
4    sp_b  12 0.7058824   7   0.875 0.7904412

I have spent some time trying to understand this discrepancy. If I have misunderstood the paper or your function please let me know. I am teaching a course next month using your package and before i confuse my students too much, I would like to know if I'm mistaken or the package has a bug.

If I am right your code could read:

RFCs <- RFCdata %>% dplyr::group_by(sp_name) %>% dplyr::summarize(FCs = sum(FCps)) %>%
  mutate(RFCs = FCs/max(FCs), FCs = NULL) %>% 
  dplyr::arrange(-RFCs)

Suggest a tutorial / training how to use the package related to ethnobotany cf. https://www.youtube.com/watch?v=QPdJNqXHxmI&lc=Ugw3YEztKdXhdtrv0mF4AaABAg

Look at possible ways to use area curves and time curves for quantitative analysis.

Read IPBES and make links (this as protocol)

Since the package is marketed as an addition to tidyverse, I think it is important all functions work with tibbles.

The function FLs() doesn't work with tibbles due to the functions use of reshape::melt(...).

The easiest fix (but not the best) is simply to replace reshape with reshape2.

The function could be rewritten to be based on tidyverse functions, to avoid problems like this.