Here, I present a workflow to analyze data from a consumer preference
-trial executed in Nigeria and Cameroon by the International Institute of
-Tropical Agriculture (IITA) under the RTBFoods project (https://rtbfoods.cirad.fr). Consumer testing was carried
-out with 1,000 participants in 2022 in Cameroon and Nigeria using the
-tricot approach [1]. Diverse consumers in
-villages, towns, and cities evaluated the overall acceptability of
-gari-eba made from 13 cassava (Manihot esculenta Crantz)
-genotypes. Apart from the overall preference of the samples, the
-following traits were evaluated for eba based on triangulated insights
-obtained through earlier surveys and participatory work in the three
-areas: Nigeria (Osun and Benue States): color, smoothness, moldability,
-stretchability, and taste. Cameroon (Littoral zone): color, odor, taste,
-firmness, and stretchability.
-
Traits in common are color, taste, and stretchability. The results of
-this study were published by Olaosebikan et al. (2023)[2], and a deeper analysis for the Nigerian
-subset, linking the consumer data to laboratory instrumental metrics,
-was published by Alamu et al. (2023)[3].
-The present vignette does not intend to replicate the results of these
-studies. Instead, it presents an alternative workflow to what was
-developed in the study. For the replication data on Olaosebikan et
-al. (2023), please visit https://github.com/AgrDataSci/rtbfoods-consumer-testing.
-For the replication data on Alamu et al. (2023), please visit https://github.com/AgrDataSci/cassava-consumer-iita/.
+
This vignette demonstrates a workflow for analyzing consumer
+preference data from decentralized trials of cassava (Manihot
+esculenta Crantz) varieties in Nigeria and Cameroon. Using the
+tricot approach [1], 1,000 participants
+evaluated gari-eba made from 13 cassava genotypes in 2022. The trial was
+implemented by the International Institute of Tropical Agriculture
+(IITA) under the RTBFoods project (https://rtbfoods.cirad.fr). Participants assessed
+overall preference and traits such as color, stretchability, and taste,
+reflecting diverse consumer priorities.
+
Building on studies by Olaosebikan et al. (2023) [2] and Alamu et al. (2023) [3], this vignette introduces an alternative
+workflow. It leverages statistical models, such as the Plackett-Luce
+model, to analyze overall preference, explore trait-specific
+performance, and account for consumer heterogeneity. By segmenting the
+data by groups like country, the analysis uncovers context-specific
+varietal performance and identifies the best varieties for specific
+groups.
+
Additionally, a weighted selection index is proposed to integrate
+multiple traits, enabling a data-driven approach to ranking varieties.
+This workflow emphasizes a consumer- and market-oriented approach,
+offering insights into plant breeding and selection strategies that
+align with local preferences and environmental contexts. The
+selection index approach is under development and open for improvements,
+suggestions and comments.
+
+
+
Read the data and select traits
+
The cassava data is a data frame with 1,000 observations
and 27 variables, which are described in the data documentation with
?cassava. This vignette will require the packages
PlackettLuce [4], ClimMobTools [5], ggplot2 [6],
and patchwork [7].
The worth map confirms the superiority of TMS6, Sape and TMEB1 across
the traits, but also presents Madame among the top varieties for color.
@@ -359,9 +357,11 @@
Participants’ heterogeneity# and get the unique values
slice_lvs=unique(slice)
-
trait_plot=list()
+# order of varieties from best to worst in the full dataset
+items_lvls=rev(names(sort(rank(coef(mod[[ov]], log =FALSE)*-1))))
+
for(iinseq_along(slice_lvs)){# fit the model also applying the slice
@@ -370,35 +370,29 @@
Participants’ heterogeneity})# plot the worth map
-trait_plot[[i]]=worth_map(mod_i,
- labels =traits)+
-labs(x ="",
- y ="",
- title =slice_lvs[i])+
-scale_fill_distiller(palette ="BrBG",
- direction =1,
- na.value ="white",
- name ="")
+trait_plot[[i]]=worth_map(mod_i,
+ labels =traits,
+ labels.order =c("Overall", "Taste", "Stretchability", "Colour"),
+ items.order =items_lvls)
-}# plot the two maps using patchworktrait_plot[[1]]+trait_plot[[2]]+plot_layout(ncol =1)
-
-
-
Selection index
-
The segmented analysis reveals contrasting results compared to the
previous analysis using the full dataset. It highlights the superiority
of TMS6 and TMEB1 in Nigeria, and Game Changer and Sape in Cameroon.
This approach provides a visual method to select the best variety
-overall for each group. To account for all traits in a data-driven
-manner when selecting varieties, I propose a selection index approach.
-This method calculates weighted estimates and derives a selection score
-that represents the overall performance of the varieties across all
-traits.
+overall for each group.
+
+
+
Selection index
+
+
To account for all traits in a data-driven manner when selecting
+varieties, I propose a selection index approach. This method calculates
+weighted estimates and derives a selection score that represents the
+overall performance of the varieties across all traits.
Suppose we have three varieties with standardized probabilities for
each trait:
Tools for Data Analysis in Experimental Agriculture
-
Methods to analyse experimental agriculture data, from data synthesis to model selection and visualisation. The package is named after W.S. Gosset aka ‘Student’, a pioneer of modern statistics in small sample experimental design and analysis.
+
+
Methods to analyse experimental agriculture data, from data synthesis to model selection and visualisation. The package is named after W.S. Gosset aka ‘Student’, a pioneer of modern statistics in small sample experimental design and analysis.
The Kendall correlation coefficient and the Effective N, which
is the equivalent N needed if all items were compared to all items.
-Can be used for significance testing.
optional, a vector to determine the order of labels
+
+
+
items.order
+
optional, a vector to determine the order of items
+
diff --git a/docs/sitemap.xml b/docs/sitemap.xml
index 2d2fcc8..d6f5eb8 100644
--- a/docs/sitemap.xml
+++ b/docs/sitemap.xml
@@ -20,6 +20,7 @@
/reference/index.html/reference/kendallTau.html/reference/kendallW.html
+/reference/kendall_permute.html/reference/kenyachoice.html/reference/likelihood_ratio.html/reference/nicabean.html
diff --git a/man/worth_map.Rd b/man/worth_map.Rd
index c9cd9e6..515e9da 100644
--- a/man/worth_map.Rd
+++ b/man/worth_map.Rd
@@ -11,7 +11,7 @@ worth_map(object, ...)
\method{worth_map}{default}(object, ...)
-\method{worth_map}{list}(object, labels, ...)
+\method{worth_map}{list}(object, labels, labels.order = NULL, items.order = NULL, ...)
worth_bar(object, ...)
}
@@ -23,6 +23,10 @@ a list objects of class \code{PlackettLuce}}
\item{...}{additional arguments passed to methods}
\item{labels}{a vector with the name of models in \var{object}}
+
+\item{labels.order}{optional, a vector to determine the order of labels}
+
+\item{items.order}{optional, a vector to determine the order of items}
}
\description{
Produces plots to highlight worth
diff --git a/vignettes/vignette-2-gari-eba-consumer-testing.R b/vignettes/vignette-2-gari-eba-consumer-testing.R
index e530735..0d3afce 100644
--- a/vignettes/vignette-2-gari-eba-consumer-testing.R
+++ b/vignettes/vignette-2-gari-eba-consumer-testing.R
@@ -91,12 +91,9 @@ ggplot(data = rel,
## ----worth, message = FALSE, eval = TRUE, echo = TRUE-------------------------
-worth_map(mod, labels = traits) +
- labs(x = "", y = "") +
- scale_fill_distiller(palette = "BrBG",
- direction = 1,
- na.value = "white",
- name = "")
+worth_map(mod,
+ labels = traits,
+ labels.order = c("Overall", "Taste", "Stretchability", "Colour"))
## ----llr, message = FALSE, eval = TRUE, echo = TRUE---------------------------
# by gender
@@ -130,9 +127,11 @@ slice = dat$country
# and get the unique values
slice_lvs = unique(slice)
-
trait_plot = list()
+# order of varieties from best to worst in the full dataset
+items_lvls = names(sort(rank(coef(mod[[ov]], log = FALSE) * -1)))
+
for (i in seq_along(slice_lvs)) {
# fit the model also applying the slice
@@ -141,17 +140,11 @@ for (i in seq_along(slice_lvs)) {
})
# plot the worth map
- trait_plot[[i]] = worth_map(mod_i,
- labels = traits) +
- labs(x = "",
- y = "",
- title = slice_lvs[i]) +
- scale_fill_distiller(palette = "BrBG",
- direction = 1,
- na.value = "white",
- name = "")
+ trait_plot[[i]] = worth_map(mod_i,
+ labels = traits,
+ labels.order = c("Overall", "Taste", "Stretchability", "Colour"),
+ items.order = items_lvls)
-
}
# plot the two maps using patchwork
diff --git a/vignettes/vignette-2-gari-eba-consumer-testing.Rmd b/vignettes/vignette-2-gari-eba-consumer-testing.Rmd
index 235d8c5..5d500b8 100755
--- a/vignettes/vignette-2-gari-eba-consumer-testing.Rmd
+++ b/vignettes/vignette-2-gari-eba-consumer-testing.Rmd
@@ -152,12 +152,9 @@ The reliability estimates show an improvement in overall preference of 32% for T
So far, we have focused solely on the overall preference. However, there are other traits to assess. The function `worth_plot()` can be used to visually analyze and compare variety performance across different traits. The values represented in a worth map are *log-worth* estimates. The function can be used with `ggplot2` functions to improve the plot.
``` {r worth, message = FALSE, eval = TRUE, echo = TRUE}
-worth_map(mod, labels = traits) +
- labs(x = "", y = "") +
- scale_fill_distiller(palette = "BrBG",
- direction = 1,
- na.value = "white",
- name = "")
+worth_map(mod,
+ labels = traits,
+ labels.order = c("Overall", "Taste", "Stretchability", "Colour"))
```
The worth map confirms the superiority of TMS6, Sape and TMEB1 across the traits, but also presents Madame among the top varieties for color. We will return to this analysis later, but for now, let us examine the data from the perspective of different groups to consider heterogeneity in the participants' evaluations, as proposed by van Etten et al. (2023) [@vanEtten2023].
@@ -204,9 +201,11 @@ slice = dat$country
# and get the unique values
slice_lvs = unique(slice)
-
trait_plot = list()
+# order of varieties from best to worst in the full dataset
+items_lvls = rev(names(sort(rank(coef(mod[[ov]], log = FALSE) * -1))))
+
for (i in seq_along(slice_lvs)) {
# fit the model also applying the slice
@@ -215,17 +214,11 @@ for (i in seq_along(slice_lvs)) {
})
# plot the worth map
- trait_plot[[i]] = worth_map(mod_i,
- labels = traits) +
- labs(x = "",
- y = "",
- title = slice_lvs[i]) +
- scale_fill_distiller(palette = "BrBG",
- direction = 1,
- na.value = "white",
- name = "")
+ trait_plot[[i]] = worth_map(mod_i,
+ labels = traits,
+ labels.order = c("Overall", "Taste", "Stretchability", "Colour"),
+ items.order = items_lvls)
-
}
# plot the two maps using patchwork
