Latrine learning: using conditional inference trees to explore how latrine conditions can predict latrine use in rural Bangladesh Public

Abstract

Global public health efforts to eliminate open defecation specifically on the Indian subcontinent have recently begun focusing on improving latrine use. In this study, we attempt to identify a latrine's likelihood of use based on observations of physical characteristics of the latrine and the surrounding premises (i.e., latrine spot-check indicators [SCIs]). Recursive partitioning algorithms, often called decision trees, are typically used in machine learning and data mining because they do not require the assumptions made by traditional regression models. Conditional inference trees (CIT) specifically apply unbiased statistical inference tests as a method of variable selection based on a priori partitioning criteria. Unlike other regression trees, the selected partitions are conditional of all other covariates in the model. In this study we measured latrine usage in rural Bangladesh in 2014 using average daily ‘likely defecation events' recorded by a motion sensing device called a passive latrine use monitor (PLUM). Using this continuous distribution, we dichotomized the measurement along its median so that we had a "Most used" group (≥ median) and a "least used" group (< median). We then employed CIT to separately predict the continuous and dichotomous forms of the outcome using 15 SCIs as independent variables. After implementing a Bonferroni correction for multiple tests of significance, the CIT analysis identified a tree with three partitions using three SCIs for the dichotomous outcome. The primary partition was the presence/absence of water for the purpose of flushing or anal cleansing with two secondary partitions being 1) the presence/absence of flies and 2) having a wet floor. The primary partition shows the strongest SCI but the secondary partitions show that a latrine with water for cleansing that does not attract flies and latrines that do not have water for this purpose but keep a dry floor draw the most use from their users. This interaction suggests a latrine's cleanliness and structural maintenance is an important indication of its use. The CIT for the continuous outcome could indicate some measurement error within the PLUMs.

Table of Contents

BACKGROUND/LITERATURE REVIEW .......................1

METHODS ...............................................................8

RESULTS .................................................................16

DISCUSSION ...........................................................19

STUDY STRENGTHS AND WEAKNESSES ...................23

FUTURE RESEARCH AND DIRECTIONS .....................25

TABLES AND FIGURES ..............................................27

REFERENCES ...........................................................37

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Epidemiology
Subfield / Discipline	Global Epidemiology - MPH & MSPH
Degree	MPH
Submission	Master's Thesis
Language	English
Research Field	Health Sciences, Public Health Psychology, Behavioral Environmental Health
Mot-clé	conditional inference tree hygiene behavior spot-check indicators latrine use decision trees recursive partitioning WASH open defecation
Committee Chair / Thesis Advisor	Clasen, Thomas, Emory University
Committee Members	Gass, Katie, Emory University Delea, Maryann , Emory University

Dernière modification

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Latrine learning: using conditional inference trees to explore how latrine conditions can predict latrine use in rural Bangladesh ()	2018-08-28 12:50:11 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions