Lectures

Statistical Biases: Confounding, Causation & Correlation

by Raywat Deonandan, PhD
(1)

Questions about the lecture
My Notes
  • Required.
Save Cancel
    Learning Material 2
    • PDF
      Slides 06 CausationBiasConfounding Epidemiology.pdf
    • PDF
      Download Lecture Overview
    Report mistake
    Transcript
    Hello and welcome to epidemiology. We're continuing our discussion of biases and things that resemble biases. And today we're talking about a particularly important kind of non-bias called a confounding, which quite excites me, because this is one of most common things that messes with studies, we'll get there though. First off, you're going to be able to identify Hawthorne and Rosenthal effects, which are common problems in randomized controlled trials, or RCT's. You're also going to be able to identify confounding variables, again that concept of confounding, which I think is so important and I think you're going to find it quite exciting as well because you are going to see it pop up in all matter of interactions you have with data and with people. And lastly, you're going to understand the concept of effect modification, which is similar to confounding, but quite distinct. So we've covered a variety of biases already, let's get into the Hawthorne and Rosenthal effects and they are combined to make something resembling placebo effect, we will talk more about placebo effect on the lecture about experiments. So Hawthorne effect is also called the observer effect and it is named after a case in the Hawthorne factory, where they discovered that when you turn the lights up in the factory the workers tended to work harder. In other words, individuals modify their behavior when they are being watched, that's the Hawthorne effect. So scientists tend to recruit patients who have a better adherence to drug therapies, who have a more likelihood of staying in a study, therefore less likelihood to have future loss to follow-up, they are special individuals, they're more likely to be good subjects. In other words, they're going to respond in a positive way more likely than other individuals not...

    About the Lecture

    The lecture Statistical Biases: Confounding, Causation & Correlation by Raywat Deonandan, PhD is from the course Statistical Biases. It contains the following chapters:

    • Statistical Biases: Confounding, Causation & Correlation
    • Hawthorne and Rosenthal Effect
    • Confounding Bias
    • Effect Modification
    • Learning Outcomes

    Included Quiz Questions

    1. Use retrospective data from cameras in store rather than having an interviewer present while the sales are taking place.
    2. Assign stores randomly amongst a variety of interviewers known to the clerks, and take data from random site visits.
    3. Advise the store clerks about the study, but obtain the data from watching security cameras in the store.
    4. Ask the store clerks to retrospectively self-report the number of instances they ask for proof of age.
    5. Ask the store clerks to note every instance they ask for proof of age over a specific period of time.
    1. Hawthorne effect
    2. Pygmalion effect
    3. Self-fulfilling prophecy
    4. Golem effect
    5. Rosenthal effect
    1. Maternal age
    2. Smoking
    3. Gender
    4. SES
    5. Any of the traditional confounders
    1. Stratify the sample by the characteristic and if the effect differs between groups than it is not a confounder.
    2. Determine if the characteristic is in the causal pathway, then it is not a confounder.
    3. It is impossible to determine if a characteristic is a confounder by statistical analysis, it must be determined prior to study design.
    4. Consult other studies of the same topic to determine if the characteristic is in the causal pathway.
    5. Stratify the sample by the characteristic and if there is no difference in direction between the groups then it is a confounder.
    1. Effect modifiers may create the illusion of an association when none exists.
    2. Effect modifiers modify the nature or direction of a real association.
    3. Effect modifiers are not a type of bias.
    4. Effect modifiers do not mask the association between two variables that are associated.
    5. Effect modifiers are described as an “interaction term” in statistics.
    1. Confirmation bias
    2. Hindsight bias
    3. Reporting bias
    4. Response bias
    5. Clustering illusion
    1. Clustering illusion
    2. Hindsight bias
    3. Information bias
    4. Confounding
    5. Confirmation bias

    Author of lecture Statistical Biases: Confounding, Causation & Correlation

     Raywat Deonandan, PhD

    Raywat Deonandan, PhD


    Customer reviews

    (1)
    4,0 of 5 stars
    5 Stars
    0
    4 Stars
    1
    3 Stars
    0
    2 Stars
    0
    1  Star
    0
     
    Liked the video
    By T?na M. on 10. December 2016 for Statistical Biases: Confounding, Causation & Correlation

    Liked the video, but wanted confounding and correlation explanations more detailed, with more explanations and etc. Had to go on youtube to see the topic more detailed. Otherwise, I LOVED the ppt slides, they are not boring, but interesting, short and east to follow.