Plant selection for ethnobotanical uses on the Amalfi Coast (Southern Italy)

Abstract

Background: Many ethnobotanical studies have investigated selection criteria for medicinal and non-medicinal plants. In this paper we test several statistical methods using different ethnobotanical datasets in order to 1) define to which extent the nature of the datasets can affect the interpretation of results; 2) determine if the selection for different plant uses is based on phylogeny, or other selection criteria.

Methods: We considered three different ethnobotanical datasets: two datasets of medicinal plants and a dataset of non-medicinal plants (handicraft production, domestic and agro-pastoral practices) and two floras of the Amalfi Coast. We performed residual analysis from linear regression, the binomial test and the Bayesian approach for calculating under-used and over-used plant families within ethnobotanical datasets. Percentages of agreement were calculated to compare the results of the analyses. We also analyzed the relationship between plant selection and phylogeny, chorology, life form and habitat using the chi-square test. Pearson's residuals for each of the significant chi-square analyses were examined for investigating alternative hypotheses of plant selection criteria.

Results: The three statistical analysis methods differed within the same dataset, and between different datasets and floras, but with some similarities. In the two medicinal datasets, only Lamiaceae was identified in both floras as an over-used family by all three statistical methods. All statistical methods in one flora agreed that Malvaceae was over-used and Poaceae under-used, but this was not found to be consistent with results of the second flora in which one statistical result was non-significant. All other families had some discrepancy in significance across methods, or floras. Significant over- or under-use was observed in only a minority of cases. The chi-square analyses were significant for phylogeny, life form and habitat. Pearson's residuals indicated a non-random selection of woody species for non-medicinal uses and an under-use of plants of temperate forests for medicinal uses.

Conclusions: Our study showed that selection criteria for plant uses (including medicinal) are not always based on phylogeny. The comparison of different statistical methods (regression, binomial and Bayesian) under different conditions led to the conclusion that the most conservative results are obtained using regression analysis.

Fig. 1

The Amalfi Coast (Italy) with boundaries of the research areas (for DTS1, DTS2 and DTS3)

Fig. 2

Proportion (and number) of species in the most represented families of the floras of the Amalfi Coast. The height of the bar represents the proportion of the flora contained in each of the top 16 families in FL1 (top panel), and the corresponding proportions of these same 16 families for FL2 appear in the bottom panel. The number of species in each family is written on the bar

Fig. 3

Medicinal plants (DTS1 and DTS2). Number of shared plant uses and taxa. Circles are proportional to the number of taxa

Fig. 4

Linear regression, binomial method and Bayesian approach applied to the three datasets (DTS1-3) in relation to FL1 and FL2. Only families with over or under-use are showed. Results show significance in over or under-uses under different statistical tests. A darker line to the right indicates significant over-use of the plant family, while a darker line to the left indicates significant under-use. When both flora indicate similar results, the darker lines of each flora overlap with the line representing FL2 longer than that of FL1

Fig. 5

Results of a power analysis that compares the minimum sample size of taxa within a family that would be required to reject the null hypothesis in favor of concluding significant under- or over-use using the binomial and Bayesian methods. The vertical dashed lines depict the average proportion for DTS1, DTS2, DTS3 for floras FL1 and FL2. The step function is the minimum sample size required to statistically define when a plant family is under- or over-used. The location where the vertical line intercepts with the step function identifies the minimum within-family number of taxa required to statistically confirm under- and over-use for that dataset and flora. The left panel shows there is no power using the binomial or the Bayesian method to detect over-use in DTS3 and FL1 (i.e., we can never conclude that families in these datasets are statistically under-used). On the other hand, the right panel shows that very few within family taxa are required to have the power to statistically conclude over-use (all datasets, all flora discussed here require two or less taxa per family), and the Bayesian method is particularly sensitive in the region of datasets DTS1, DTS2 in FL2

Fig. 6

Percentages of agreement and discordance among the results obtained using different methods (linear regression, binomial method and Bayesian approach), with fixed floras. Agreement is when the significance is the same (both not significant, both significant and in the same direction)

Fig. 7

Percentages of agreement and discordance using floras FL1 and FL2 among the results obtained using the three statistical methods

Fig. 8

Number of under-used or over-used families using the different methods (linear regression, binomial method and Bayesian approach), floras (FL1 and FL2) and datasets (DTS1, DTS2, DTS3)

Fig. 9

Plots of Pearson’s residuals for phylogeny, life form and habitat for DTS1, DTS2 and DTS3