nerc.ac.uk

Controlling the marginal false discovery rate in inferences from a soil dataset with α -investment

Lark, R.M.. 2017 Controlling the marginal false discovery rate in inferences from a soil dataset with α -investment. European Journal of Soil Science, 68 (2). 221-234. 10.1111/ejss.12413

Before downloading, please read NORA policies.
[img]
Preview
Text (Open Access Paper)
ejss12413.pdf - Published Version
Available under License Creative Commons Attribution 4.0.

Download (514kB) | Preview

Abstract/Summary

Large datasets on soil provide a temptation to search for relations between variables and then to model and make inferences about them with statistical methods more properly used to test preplanned hypotheses on data from designed experiments or sample surveys. The control of family-wise error rate (FWER) is one way to improve the robustness of inferences from tests of multiple hypotheses. In its simplest form, hypothesis testing with FWER control lacks statistical power. The α-investment approach to controlling the marginal false discovery rate is one method proposed to improve statistical power. In this paper I outline the α-investment approach and then demonstrate it in the analysis of a dataset on the rate of CO2 emission from incubated intact cores of soil from a transect over Cretaceous rocks in eastern England. Hypotheses are advanced after considering the literature and examining relations among the available soil variables that might be proposed as explanatory factors for the variation of CO2 emissions. They are then tested in sequence with α-investment, such that the rejection of null hypotheses increases the power to reject later ones, while controlling the overall marginal false discovery rate at a specified value. This paper illustrates the use of α-investment to test a multiple set of hypotheses on a soil dataset; statistical power is improved by ordering the sequence of hypotheses on the basis of process knowledge. The approach could be useful in other areas of soil science where covariates must be selected for predictive statistical models, notably in the development of pedotransfer functions and in digital soil mapping. Highlights α-investment controls marginal false discovery rate in statistical inference. Hypotheses were advanced about soil factors that affect CO2 emission from soil. These hypotheses were tested in sequence with control of marginal false discovery rate. Soil properties, land use and parent material were significant predictors.

Item Type: Publication - Article
Digital Object Identifier (DOI): 10.1111/ejss.12413
ISSN: 13510754
Date made live: 07 Mar 2017 09:15 +0 (UTC)
URI: http://nora.nerc.ac.uk/id/eprint/516456

Actions (login required)

View Item View Item

Document Downloads

Downloads for past 30 days

Downloads per month over past year

More statistics for this item...