** Key Chapter 12 Review Questions Key **

�12.1 Introduction

  1. Systematic error and random error. 
  2. Parameters are error-free quantifications.  Estimates are error-prone statistics.
  3. Estimators have hats. Parameters are �hatless.�
  4. Bias
  5. Imprecision.
  6. Valid
  7. Imprecise
  8. Random error is balanced (equal number of overestimates and underestimates), is sample size dependent (less random error in large samples), and can be dealt with via laws of probability.

�12.2 Random Error 

  1. False. Probability models only address random error.
  2. False. Both subjective and objective probabilities are based on an underlying concept of expected frequencies. In addition, they obey the same mathematical laws.
  3. Confidence intervals and hypothesis tests 
  4. Random error would be eliminated. Increasing the sample size has no effect on systematic error.
  5. True. Random error causes our estimates to be imprecise. 

�12.3 Systematic Error

  1. Selection, information, confounding
  2. False. Nondifferential misclassification biases measures of effect toward the null, or in some instances, not at all.
  3. False. Bias away from the  null overstates risk. 
  4. Confounding is a distortion in a measure of effect brought about by extraneous ("lurking") factors.
  5. (a) Associated with the exposure (b) Independent risk factor for disease (c) not intermediate in the causal pathway
  6. Confundere = to mix-up
  7. Berkson's bias, also know as "Hospital admission rate"
  8. No. It will tend to have the same amount of bias. (It will have less random error, however.)
  9. Recall bias
  10. Information bias
  11. When it is not associated with the exposure
  12. This question from Rothman is thought provoking. Clearly, smoking causes lung cancer in both men and women. Given the source of the question, one can hypothesize that a response should address causal mechanisms and causal complements in the populations. It the populations have similar causal mechanisms, then we can generalize a study to a population whose characteristics differ from those in the study population. Otherwise, we cannot.
  13. It could be considered confounding if the medical indication or the severity of the condition can be measured and adjusted for in the analysis, or it might be considered uncontrolled confounding or selection bias because it acts like a confounder but is a consequence of being "selected" to get the drug.