+ Reply to Thread
Results 1 to 11 of 11

Thread: Establishing Causality

  1. #1
    Points: 220, Level: 4
    Level completed: 40%, Points required for next Level: 30

    Posts
    4
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Establishing Causality




    Hi All,
    I'm new to this forum and have a very basic question.

    How do we establish causality after establishing association?

    Do we have any methods which are generally accepted across the field and how reliable are they?

    If you could point me in the right direction towards either some kind of MOOC or a book it would be of great help to me.

    Thanks

  2. #2
    Omega Contributor
    Points: 38,334, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,185 Times in 1,146 Posts

    Re: Establishing Causality

    Will you be working with observational data or experimental data with possible randomization?

    The best straightforward book without too much formulae is the counterfactual book by Morgan and winship.
    Stop cowardice, ban guns!

  3. The Following 2 Users Say Thank You to hlsmith For This Useful Post:

    katxt (05-23-2017), shabd_a (05-23-2017)

  4. #3
    Points: 220, Level: 4
    Level completed: 40%, Points required for next Level: 30

    Posts
    4
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Establishing Causality

    Thanks hlsmith for the reply.

    I will be dealing with observational data.Since the data will be related to macroeconomics, it won't be possible to get experimental data.

    Thanks for the book recommendation.Will surely check it out.

  5. #4
    Omega Contributor
    Points: 38,334, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,185 Times in 1,146 Posts

    Re: Establishing Causality

    Great. I placed some links below, since yesterday I was using a small device that made attaching links difficult. Once you start reading causal literature you will quickly pick up who the big names in the area are and I recommend reading their books, papers, and blogs as well. With economic you will get exposure to instrumental variables, interrupted time series, regression discontinuity, etc.


    Let me know if you have any questions when moving forward in that this is an area I have great interest in. The big thing is addressing non-random allocation of exposures and potentially related known and unknown confounders.






    https://www.amazon.com/Counterfactua...9PZFGJ40ZE5RNX


    https://www.amazon.com/Causal-Infere...F57GT39NMPBSGG
    Stop cowardice, ban guns!

  6. #5
    Points: 220, Level: 4
    Level completed: 40%, Points required for next Level: 30

    Posts
    4
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Establishing Causality

    Thanks hlsmith.

    I have only the basic level understanding of statistics at this point and trying to gain more as I am interested in it and going on some personal projects at work.Will surely be checking out the recommended books and post if I have any questions.Do you think there are some other concepts I should be checking out before going through these books?

    Thanks for your help once again.

  7. #6
    Omega Contributor
    Points: 38,334, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,185 Times in 1,146 Posts

    Re: Establishing Causality

    Well in the bigger picture, most causal approaches can be very similar to the standard statistical approaches, but they require a few more assumptions in order to make conclusions/inferences causal. The use of propensity scores is also very common in these approaches along with counterfactuals concepts.


    Lastly, whenever I am trying to convey a more causal relationship, I try to refer to related/applicable concepts listed in the Bradford-Hill Criteria (which were originally focused on more infectious diseases, so not all are applicable - though they give a framework of generalities to consider in designing and conducting research).
    Stop cowardice, ban guns!

  8. #7
    Points: 3,506, Level: 37
    Level completed: 4%, Points required for next Level: 144

    Posts
    129
    Thanks
    3
    Thanked 26 Times in 26 Posts

    Re: Establishing Causality

    I think I would argue that the answer to the original question of how to establish causality is that with observational designs you really can't. You develop evidence and deal with counterarguments, but with rare exceptions actual proof of causality is not within the scope of that kind of research.

  9. #8
    TS Contributor
    Points: 18,889, Level: 87
    Level completed: 8%, Points required for next Level: 461
    CowboyBear's Avatar
    Location
    New Zealand
    Posts
    2,062
    Thanks
    121
    Thanked 427 Times in 328 Posts

    Re: Establishing Causality

    Quote Originally Posted by EdGr View Post
    I think I would argue that the answer to the original question of how to establish causality is that with observational designs you really can't. You develop evidence and deal with counterarguments, but with rare exceptions actual proof of causality is not within the scope of that kind of research.
    True, but you can't prove causality with experimental research either - proof implies absolute certainty, which we can never reach when dealing with contingent claims about the world. (Hence the need for using statistics and probability!) Experimental research does allow for less uncertainty about causal claims though.
    Matt aka CB | twitter.com/matthewmatix

  10. #9
    Points: 220, Level: 4
    Level completed: 40%, Points required for next Level: 30

    Posts
    4
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Establishing Causality

    Hi EdGr. Thanks for your response.Could you elaborate more on why it's impossible.

  11. #10
    Omega Contributor
    Points: 38,334, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,185 Times in 1,146 Posts

    Re: Establishing Causality

    This may be of interest to all, the ASCO is now officially championing the contribution of observational research.


    http://ascopubs.org/doi/pdf/10.1200/JCO.2017.72.6414
    Stop cowardice, ban guns!

  12. #11
    Points: 3,506, Level: 37
    Level completed: 4%, Points required for next Level: 144

    Posts
    129
    Thanks
    3
    Thanked 26 Times in 26 Posts

    Re: Establishing Causality


    "Hi EdGr. Thanks for your response.Could you elaborate more on why it's impossible. "

    As Cowboybear said, all things are relative. But with observational research, we do not have full control over third variables. There can always be something else we didn't consider differing between groups that is the real cause. There are lots of ways to reduce that risk, including careful research design, attention to confounders, appropriate models, and so on. Also, as our knowledge advances, we should find combinations of predictors with such high associations that confounding is unlikely. For example, say people who take a certain drug seem to have less recurrence of cancer (in observational studies). Then we notice that in the subset of cancer patients with mutation X, none of them with the drug recurred, and ALL without the drug recurred. Wow... pretty convincing.

    Randomized trials provide stronger evidence when they are well done because they better control confounding.

    As evidence for the difficulty (not impossibility, perhaps) consider all the well-done observational studies of diet and disease that seem to find a different thing each time a new study comes out!

    If you try to publish an observational study and make statements implying that your results "show causality" I think you will have editors objecting. On the other hand, a well-done randomized trial can get away with saying that it "showed" treatment X is better than treatment Y.

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats