5: Chapter 6

Course Information

• Course title: BMS 511 Biostats & Statistical Analysis
• Chapter title: Samples and Observational Studies
• Instructor: Guang Xu, PhD, MPH
• Affiliation: Assistant Professor of Biostatistics and Public Health, College of Osteopathic Medicine, Marian University
• Textbook Copyright: © 2018 W. H. Freeman and Company

Previous Learning Objectives

• Least-squares regression line
• Facts about least-squares regression
• Outliers and influential observations
• Working with logarithm transformations
• Cautions about correlation and regression
• Association does not imply causation

Learning Objectives

• Determine samples and observational studies
• Observational study versus experiment
• Population versus sample
• Randomness and bias
• Simple random sample (SRS)
• Other probability samples
• Sample surveys
• Comparative observational studies

Observational vs Experimental Studies (1 of 2)

• Observational study: Recording data on individuals without influencing their responses.
• Example: 1992, medical organizations recommended estrogen after menopause, due to perceived reduction in heart attack risk (35% to 50%).

Observational vs Experimental Studies (2 of 2)

• Experimental study: Deliberately imposing a treatment and recording responses. Influential factors are controlled.
• Example: By 2002, studies found hormone replacement did not lower the risk of heart attacks. Studies randomly assigned women to hormone replacement or a placebo.

Observational Study or Experiment? (1 of 2)

• 2013 Gallup study investigated phrasing effect on American opinions regarding physician-assisted suicide.
• Telephone interviews conducted with 1,535 adults.
• Random assignment used, with 719 hearing Form A and 816 hearing Form B.
• Form A: "When a person has a disease that cannot be cured, do you think doctors should be allowed by law to end the patient's life by some painless means if the patient and his or her family request it?"

Observational Study or Experiment? (2 of 2)

• Form B: "When a person has a disease that cannot be cured and is living in severe pain, do you think doctors should or should not be allowed by law to assist the patient to commit suicide if the patient requests it?"
• 70% of those given Form A answered "should be allowed", compared to only 51% of those given Form B.
• This is an observational study, not an experiment.

Confounding

• Two variables are confounded when their effects on a response variable cannot be distinguished.
• Observational studies often fail to demonstrate clear causal conclusions because explanatory variables are confounded with lurking variables.
• Example: Wine vs. beer (explanatory) vs. health (response) vs. diet and lifestyle (lurking)

Population vs Sample

• Population: Entire group of individuals of interest, but often not directly assessed.
• Parameter: Number that summarizes a population characteristic.
• Sample: Part of the population that is examined; data are collected from this part.
• Statistic: Number that summarizes a sample characteristic.

Role of Randomness in Sampling

• Probability sampling: Random selection of individuals or units. Unbiased sampling process.

Bias

• Bias: Systematic tendency for a study to favor certain outcomes.
• Illustrative diagram shows how different samples can vary.

Examples of Bad Sampling (1 of 2)

• Ann Landers response summary: 70% of (~10,000) parents said having kids wasn't worth it.
• Random sample showed 91% of parents would have kids again.

Examples of Bad Sampling (2 of 2)

• Would survey responses differ based on the location (e.g., university vs church parking lots) of the interviewee selection?

The Simple Random Sample (SRS)

• Simple random sample (SRS): Made of randomly selected individuals.
• Each individual has the same probability of being in the sample.
• All possible samples of a given size have the same chance of being selected.
• Methods for creating an SRS:
  • Drawing from a hat
  • Using a table of random numbers
  • Using software that generates random numbers

Choosing a Simple Random Sample (1 of 2)

• Selecting a sample of 5 students from a class of 20.
• List students and assign a number to each.
• Use a random number table or software to choose 5 numbers.
• Individuals corresponding to the selected numbers form the SRS

Choosing a Simple Random Sample (2 of 2)

• A list of 20 students, numbered for easier selection.
• Provided as a reference table

Other Probability Samples (1 of 2)

• Stratified random sample: Ensures sample includes known percentages of different types (strata).
• Example: A study using Medco's de-identified database to ensure age and gender match the demographics of the customer base.
• Multistage sampling: Selecting final sample through stages/samples within samples.

Other Probability Samples (2 of 2)

• Example: National Youth Tobacco Survey uses a procedure for generating a nationally representative sample of 6-12 graders.
  • Counties as Primary Sampling Units (PSU)
  • Schools within each PSU
  • Classes within each school.

Sample Surveys

• Sample survey: Observational study using random sample of population to collect data.
• Methods examples:
  • Opinion polls
  • Voter registries/phone numbers for selection
  • Incidence/prevalence measures for diseases/lifestyles (stratified or multistage samples)

Some Survey Challenges

• Undercoverage: Parts of population systematically omitted.
• Nonresponse: Some participants don't answer.
• Question wording effects: Bias and/or leading questions.
• Response bias: Lying or forgetting, especially on sensitive issues. (In-person vs phone/online can affect this.)

How Bad is Nonresponse in Surveys? (1 of 2)

• Census Bureau's American Community Survey (ACS): ~2.5% nonresponse rate.
• University of Chicago’s General Social Survey (GSS): ~30% nonresponse rate.
• Pew Research Center’s surveys up to ~90% of desired sample size in 2012.
• Private polling firms: ~90% nonresponse rate in 2002.

How Bad is Nonresponse in Surveys? (2 of 2)

• Data on contact/response rates in sample surveys from 1997 to 2012, showing decreasing trend.

Wording Effects (1 of 2)

• 2013 Gallup study examining phrasing effects on American opinions regarding physician-assisted suicide.
• Random assignment of questions.
• Form A: "Do you think doctors should be allowed by law to end the patient's life by some painless means if requested by patient or family?"

Wording Effects (2 of 2)

• Form B: "When suffering with a terminal illness and severe pain, do you think doctors should assist a patient with suicide if requested?"
• Question wording produced a substantial difference in response.

Example of Response Bias (1 of 2)

• Gallup's annual consumption poll (July 9-12, 2012), shows response bias in reported cigarette consumption.
• Potential that respondents may present themselves in the best possible light (understating vs accurately reporting smoking behavior).

Example of Response Bias (2 of 2)

• Data from the National Health Examination Survey (2001-2006) show self-reported height vs measured height, as well as weight (kg) vs. reported weight (kg) for males and females. Self-reported values tend to overestimate height and weight for men, and overestimate height for women. Self-report values tend to underestimate women's weight.

Comparative Observational Studies

• Case-control studies: Start with random samples with opposite outcomes, to look for factors in the subject's past.
  Case patients vs Controls
• Good for rare conditions; challenging to select appropriate control groups.
• Cohort studies: Individuals with similar demographics tracked over long periods to assess factors over time.
• Good for common conditions, expansive. Cohort tracking is time consuming and can be expensive.

A Case-Control Study Example (1 of 2)

• Aflatoxicosis epidemics: Aflatoxins (harmful substances) produced by a fungus found in damaged crops.
• 2004 Outbreak in Kenya resulted in over 300 liver failure cases.
• Case patients and health controls were interviewed for maize storing practices.

A Case-Control Study Example (2 of 2)

• Patients were randomly chosen from a hospital list of unexplained acute jaundice patients.
• Controls were similar to patients, randomly selected.

Examples of Cohort Studies (1 of 2)

• Nurses' Health Study (1976-present): Monitoring >100,000 registered nurses.
• Biennial questionnaires about diseases and health.
• High response rate (~90% per survey).

Examples of Cohort Studies (2 of 2)

• 2007 memory loss study: Assessing 20,000 women over 70 years old.
• Telephone interviews every 2 years to determine memory functions.
• Study found correlation between walking and possible memory benefits (observational study, so cannot prove causation).

Application of SPSS: Least-Square Regression Line

• Details on using SPSS (statistical software) to create linear regressions.
• Instructions on generating graphs demonstrating data and fitting regression lines. Different methods for creating linear regressions via SPSS were demonstrated.
• Example data and results will show how plots, trendlines and fit lines are generated.

Test your knowledge on key concepts in research methods and statistics with this quiz. Explore topics like sampling techniques, bias in studies, and the differences between observational and experimental studies. Perfect for students seeking to deepen their understanding of statistical research principles.