Masters Class 2
Proposal Development
Research Question and Hypotheses

September 27, 2004

Proposal development

How to begin?

Begin at the beginning and go on till you come to the end; then stop. --Lewis Carroll

In the beginning, no data were available . . .

2 guest speakers

Anne Paxton, DrPH

Assistant Professor of Population & Family Health

EXPERTISE:

Design, monitoring and evaluation of public health research and service programs in developing countries.

Adaptation of epidemiologic methodologies to resource-poor settings.

Areas of interest include women's reproductive health, prevention of maternal mortality, nutrition in pregnancy, trachoma prevention and control and social epidemiology.

[email protected]

Alfred I. Neugut, MD, PhD

Professor of Medicine and Epidemiology

Head of Cancer prevention and Control for the Herbert Irving Comprehensive Cancer Center

Co-Director of the Cancer Prevention Center of New York Presbyterian Hospital

PI, NCI-funded Training Program in Cancer Epidemiology, Biostatistics, and Environmental Health Sciences

[email protected]

Problem to be addressed, or research question

1-2 sentences

Exposure(s)

Outcome(s)

Person, place, and time:

E.g., association between taking P9419 and getting a master’s thesis proposal approved, among students who entered the master’s program in epidemiology at the Mailman SPH in 2000-02.

Where do research questions come from?

Start with outcomes (e.g., cancer, infectious disease) ® think about risk factors or exposures

Start with exposures ( e.g., environmental, nutritional) ® think about outcomes

BTW, intervention (clinical trial) = exposure

Reading the literature

Interacting with faculty, classmates, coworkers, etc.

Datasets

Hypotheses (1-3) must:

Be closely related to the research question

Be independent from one another, e.g.,

Risk for lung cancer is higher in smokers than in nonsmokers.

Risk for lung cancer is higher in individuals exposed to radiation than in unexposed individuals.

Include the nature and direction of the association

Smokers develop lung cancer at a younger age than do nonsmokers

Directionality¹causality

Cannot evaluate causality based on a single data analysis (except some RCTs).

Can assess correlations, dose/response, likelihood of outcome given exposure vs. no exposure, or high level of exposure vs low level of exposure

What does a dataset contain?

Variables

Demographics (age, sex, etc.)

Risk factors X₁, X₂, X₃ . . .

Outcomes Y₁, Y₂, Y₃ . . .

Values

Continuous

Categorical

Things you need to know about your dataset

What variables will be available to be analyzed?

What do they mean?

Where do they come from?

Questionnaires

Log sheets or abstracting forms

Data dictionary – variable names and meanings (e.g., SMOK=Did you ever smoke more than 100 cigarettes in a year?)

Codebook (e.g, 1=Yes, 2=No)

Be specific!

Research question ~ the association of psychosocial factors with asthma in 4-year-old children attending Head Start in New York in 2004.

Hypothesis ~ maternal risk for depression based on CESD score is associated with number of ED visits for asthma (among the children), controlling for age, sex, ethnicity . . .

Hypothesis formulation is iterative.

Review literature

Talk with readers/coworkers, etc.

Dataset

Go back to literature

Go back to readers

Go back to dataset

Good hypotheses make good methods.

Hypothesis must be testable. (Exploratory/pilot analyses are OK as long as you acknowledge limitations.)

Think in terms of regression model:

Y=b₁X₁+ b₂X₂+ b₃X₃+E

Think about directionality:

Y when X

Y¯ when X

Groups

Chronic disease (includes aging, cardio, neuro, and pulmonary)

Psychosocial (includes violence/trauma, juvenile crime, etc.)

Cancer

HIV/AIDS

Other infectious disease

Other


	How to begin?
	Begin at the beginning and go on till you come to the end; then stop. --Lewis Carroll
	In the beginning, no data were available . . .
	2 guest speakers


	Assistant Professor of Population & Family Health
	EXPERTISE:
		Design, monitoring and evaluation of public health research and service programs in developing countries.
		Adaptation of epidemiologic methodologies to resource-poor settings.
		Areas of interest include women's reproductive health, prevention of maternal mortality, nutrition in pregnancy, trachoma prevention and control and social epidemiology.
	[email protected]


	Professor of Medicine and Epidemiology
	Head of Cancer prevention and Control for the Herbert Irving Comprehensive Cancer Center
	Co-Director of the Cancer Prevention Center of New York Presbyterian Hospital
	PI, NCI-funded Training Program in Cancer Epidemiology, Biostatistics, and Environmental Health Sciences
	[email protected]


	1-2 sentences
	Exposure(s)
	Outcome(s)
	Person, place, and time:
		E.g., association between taking P9419 and getting a master’s thesis proposal approved, among students who entered the master’s program in epidemiology at the Mailman SPH in 2000-02.


	Start with outcomes (e.g., cancer, infectious disease) ® think about risk factors or exposures
	Start with exposures ( e.g., environmental, nutritional) ® think about outcomes
	BTW, intervention (clinical trial) = exposure
	Reading the literature
	Interacting with faculty, classmates, coworkers, etc.
	Datasets


	Be closely related to the research question
	Be independent from one another, e.g.,
		Risk for lung cancer is higher in smokers than in nonsmokers.
		Risk for lung cancer is higher in individuals exposed to radiation than in unexposed individuals.
	Include the nature and direction of the association
		Smokers develop lung cancer at a younger age than do nonsmokers


	Cannot evaluate causality based on a single data analysis (except some RCTs).
	Can assess correlations, dose/response, likelihood of outcome given exposure vs. no exposure, or high level of exposure vs low level of exposure


	Variables
		Demographics (age, sex, etc.)
		Risk factors X₁, X₂, X₃ . . .
		Outcomes Y₁, Y₂, Y₃ . . .
	Values
		Continuous
		Categorical


	What variables will be available to be analyzed?
	What do they mean?
	Where do they come from?
		Questionnaires
		Log sheets or abstracting forms
	Data dictionary – variable names and meanings (e.g., SMOK=Did you ever smoke more than 100 cigarettes in a year?)
	Codebook (e.g, 1=Yes, 2=No)


	Research question ~ the association of psychosocial factors with asthma in 4-year-old children attending Head Start in New York in 2004.
	Hypothesis ~ maternal risk for depression based on CESD score is associated with number of ED visits for asthma (among the children), controlling for age, sex, ethnicity . . .


	Review literature
	Talk with readers/coworkers, etc.
	Dataset
	Go back to literature
	Go back to readers
	Go back to dataset


	Hypothesis must be testable. (Exploratory/pilot analyses are OK as long as you acknowledge limitations.)
	Think in terms of regression model:
		Y=b₁X₁+ b₂X₂+ b₃X₃+E
	Think about directionality:
		Y when X
		Y¯ when X


	Chronic disease (includes aging, cardio, neuro, and pulmonary)
	Psychosocial (includes violence/trauma, juvenile crime, etc.)
	Cancer
	HIV/AIDS
	Other infectious disease
	Other