+ Reply to Thread
Results 1 to 4 of 4

Thread: Help: regression analysis for categorical/binary exposure

  1. #1
    Points: 11, Level: 1
    Level completed: 21%, Points required for next Level: 39

    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Help: regression analysis for categorical/binary exposure




    Cell biologist in need of stats help!

    I am using stata to analyse my data and have two questions:

    1. I have a continuous outcome (hippocampal volume) and exposures grouped either as categorical (4 levels) or binary. Is linear regression an appropriate analysis for this data?

    2. When I construct basic scatter plots between my exposure and outcome, there looks to be no or very little correlation. Am I right in assuming you need to see some kind of relationship to conduct linear regression? In that case, is there anything I can do to my data to allow me to analyse it?

    I have many covariates that need to be considered- I am not sure if going straight ahead and running models to adjust for age, sex etc would be useful.... All online sources seem to imply you must see a linear relationship between E -> O to conduct linear regression :/

    Any advice would be gratefully received
    Last edited by Leucine; 01-28-2017 at 07:09 PM.

  2. #2
    Points: 842, Level: 15
    Level completed: 43%, Points required for next Level: 58

    Posts
    138
    Thanks
    13
    Thanked 39 Times in 36 Posts

    Re: Help: regression analysis for categorical/binary exposure

    Quote Originally Posted by Leucine View Post
    Cell biologist in need of stats help!

    I am using stata to analyse my data and have two questions:

    1. I have a continuous outcome (hippocampal volume) and exposures grouped either as categorical (4 levels) or binary. Is linear regression an appropriate analysis for this data?

    2. When I construct basic scatter plots between my exposure and outcome, there looks to be no or very little correlation. Am I right in assuming you need to see some kind of relationship to conduct linear regression? In that case, is there anything I can do to my data to allow me to analyse it?

    Any advice would be gratefully received
    1. MLR sounds like it would be an option, but give us some more information to help you out!
    2. Have you tried just looking at the means (without any formal tests) for each of the levels? This can sometimes give you a big picture idea of differences between the groups.

    If you could tell us your sample size and give us a full listing of the variables you need to work with (including their type and marking the ones that are of particular interest to your research question as well as those included just to mitigate confounding), that would be helpful.

  3. #3
    Points: 11, Level: 1
    Level completed: 21%, Points required for next Level: 39

    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Help: regression analysis for categorical/binary exposure

    Thank you for your quick reply! Some extra details:

    I have a sample size of 320, age range 15-17.
    I am looking at the effect of smoking on hippocampal volume- so have lots of different variables for smoking including binary and categorical (but no continuous measures)

    Other variables of interest as an outcome: IQ (continuous), total brain volume (continuous)

    Potential effect mediators and confounders I have marked as of interest: sex (binary), age (cont), SES (catergorical), caffeine use (categorical), alcohol use(binary, categorical), recreational drug use (binary), depression (binary: high/low) and anxiety (binary:high/low)

    Have not tried looking at the means of each level but that makes a lot of sense- will have a go at that now (so tempting to jump into 'big' analysis when you're not very familiar with stats..)

  4. #4
    Points: 842, Level: 15
    Level completed: 43%, Points required for next Level: 58

    Posts
    138
    Thanks
    13
    Thanked 39 Times in 36 Posts

    Re: Help: regression analysis for categorical/binary exposure


    Quote Originally Posted by Leucine View Post
    Thank you for your quick reply! Some extra details:

    I have a sample size of 320, age range 15-17.
    I am looking at the effect of smoking on hippocampal volume- so have lots of different variables for smoking including binary and categorical (but no continuous measures)
    The narrow age range is going to put a firm collar on your inferences (i.e. you can't really project this to anyone outside this narrow range give).

    Quote Originally Posted by Leucine View Post
    Other variables of interest as an outcome: IQ (continuous), total brain volume (continuous)
    Do you have any theory or prior research to suggest that either of these two are correlated with hippocampal volume?

    Quote Originally Posted by Leucine View Post
    Potential effect mediators and confounders I have marked as of interest: sex (binary), age (cont), SES (catergorical), caffeine use (categorical), alcohol use(binary, categorical), recreational drug use (binary), depression (binary: high/low) and anxiety (binary:high/low)
    How many levels are in each of these polychotomous variables? What are the levels for each? Are depression and anxiety self-determined or clinical diagnoses?

    Given that there are many listed here, and the potential for an unwieldy model exits, I would think carefully about each of the potential interactions (effect mediators). If you don't have solid reasoning to support that relationship, it may be worth excluding some that aren't logical until we know how large the model ends up being.

    If there is a good literature support for some of these variables with any of the outcomes, I would include those variables as well (think of alcohol on brain development in adolescents, for example).

    Quote Originally Posted by Leucine View Post
    Have not tried looking at the means of each level but that makes a lot of sense- will have a go at that now (so tempting to jump into 'big' analysis when you're not very familiar with stats..)
    It's always exciting to jump right in, but this is an often overlooked area to start. Exploring the data and getting a feel for patterns can help in determining the model or what actually can or should be included.

+ 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