Abstract
BACKGROUND: There are handful hypothesis-driven ethnobotanical studies in Nepal. In this study, we tested the non-random medicinal plant selection hypothesis using national- and community-level datasets through three different types of regression: linear model with raw data, linear model with log-transformed data and negative binomial model. METHODS: For each of these model, we identified over-utilized families as those with highest positive Studentized residuals and underutilized families with highest negative Studentized residuals. The national-level data were collected from online databases and available literature while the community-level data were collected from Baitadi and Darchula districts. RESULTS: Both dataset showed larger variance (national dataset mean 6.51 < variance 156.31, community dataset mean 1.16 < variance 2.38). All three types of regression were important to determine the medicinal plant species selection and use differences among the total plant families, although negative binomial regression was most useful. The negative binomial showed a positive nonlinear relationship between total plant family size and number of medicinal species per family for the national dataset (β1 = 0.0160 ± 0.0009, Z1 = 16.59, p < 0.00001, AIC1 = 1181), and with similar slope and stronger performance for the community dataset (β2 = 0.1747 ± 0.0199, Z2 = 8.76, p < 0.00001, AIC2 = 270.78). Moraceae and Euphorbiaceae were found over-utilized while Rosaceae, Cyperaceae and Caryophyllaceae were recorded as underutilized. CONCLUSIONS: As our datasets showed larger variance, negative binomial regression was found the most useful for testing non-random medicinal plant selection hypothesis. The predictions made by non-random selection of medicinal plants hypothesis holds true for community-level studies. The identification of over-utilized families is the first step toward sustainable conservation of plant resources and it provides a baseline for pharmacological research that might be leading to drug discovery.