+ Reply to Thread
Results 1 to 4 of 4

Thread: Ordinal Regression vs Spearmans Rank Correlation

  1. #1
    Points: 38, Level: 1
    Level completed: 76%, Points required for next Level: 12

    Location
    UK
    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Ordinal Regression vs Spearmans Rank Correlation




    Hello,

    My name is William, I am currently at my last year at Newcastle University, UK carrying out research for my dissertation on how social media can effect levels of trust in brands. I have sent out a questionnaire and got 383 valid responses. I have developed a theory (cycle) that as valued content is posted on social media it is engaged upon, higher engagement leads to stronger relationships with brands, stronger relationships leads to more online recommendation, people are recommending your brand and your content. Cycle returns to the beginning

    Valued Content —> Engagement —> Relationship —> Recommendation ---> back to Content

    Hypotheses
    H1, Consumer engagement and interaction is positively influenced by valued content posted
    H2, Brand-consumer relationships are positively influenced by engagement
    H3, Brand page/brand community recommendations are positively influenced by online consumer-brand relationship
    H4, Increased brand page recommendations results in content that is valued higher
    H5, Brand trust is positively influenced by positive correlations in H1, H2, H3 and H4

    Each concept is indicated by three 5 point Likert items that are coded (1-5) and added together to provide a numerical measure for each concept (out of 15).

    My question is for the analysis, I am querying whether to carry out ordinal regression with path analysis on SPSS however my stats skills are limited. Or on the other hand to use Spearmans Rank Correlation between each concept?

    I keep reading in the literature that ordinal data from Likert items and scales is tricky to handle when trying to carry out regression analysis, parametric tests and such as there is no way to tell that the difference between strongly agree and agree is the same as agree and impartial etc,.

    At the end of that day I am asking whether it is worth my time to carry out and learn ordinal regression or is Spearmans Rank a good enough analysis, I realise that my dissertation won’t change the world but I have worked hard at university and would like to get a good degree.

    Many thanks

  2. #2
    Points: 3,457, Level: 36
    Level completed: 72%, Points required for next Level: 43

    Posts
    156
    Thanks
    42
    Thanked 43 Times in 40 Posts

    Re: Ordinal Regression vs Spearmans Rank Correlation

    Ordinal logistic regression differs from Spearman rank correlation in that regression explicitly defines one variable as dependent (or outcome) and one as independent (or predictor). This seems to be close to what you want.

    Ordinal logistic regression also allows you to have more than two variables in the relationship. You don't explicitly say that you want to do this, but it seems likely that you do.

    However, if you have added together three Likert scales to form variables, then you have already assumed that the Likert scales are interval level - you can't add ordinal variables. This sort of thing is done all the time. But, once you add 3 five point Likert scales, you get a scale that goes from 3 to 15 and you might be able to use OLS regression (that is, "regular" regression). Technically, that requires that the dependent variable be continuous, but this is also violated all the time. Many variables are not measured on a continuous scale (one could argue that NO variables are), it's just a question of how many levels you need.

  3. #3
    Points: 38, Level: 1
    Level completed: 76%, Points required for next Level: 12

    Location
    UK
    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Ordinal Regression vs Spearmans Rank Correlation

    Quote Originally Posted by PeterFlom View Post
    Ordinal logistic regression differs from Spearman rank correlation in that regression explicitly defines one variable as dependent (or outcome) and one as independent (or predictor). This seems to be close to what you want.

    Ordinal logistic regression also allows you to have more than two variables in the relationship. You don't explicitly say that you want to do this, but it seems likely that you do.

    However, if you have added together three Likert scales to form variables, then you have already assumed that the Likert scales are interval level - you can't add ordinal variables. This sort of thing is done all the time. But, once you add 3 five point Likert scales, you get a scale that goes from 3 to 15 and you might be able to use OLS regression (that is, "regular" regression). Technically, that requires that the dependent variable be continuous, but this is also violated all the time. Many variables are not measured on a continuous scale (one could argue that NO variables are), it's just a question of how many levels you need.
    Thank you very much, it seems that Ordinal Regression is what I am after

    However if adding Likert items together is not the proper way of constructing a variable despite the fact that it is done all the time, what is the most appropriate way of constructing a variable from 3 Likert item questions?

  4. #4
    Points: 3,457, Level: 36
    Level completed: 72%, Points required for next Level: 43

    Posts
    156
    Thanks
    42
    Thanked 43 Times in 40 Posts

    Re: Ordinal Regression vs Spearmans Rank Correlation


    The problem is that, while it is technically wrong, it is quite likely correct on a practical level.

    Stevens' scale of "nominal, ordinal, interval, ratio" has become enshrined. But, really, lots of scales are either not classifiable on this scheme or are in-between two categories. Likert scales are technically ordinal. But that would mean that we could use, e.g.

    Strongly disagree = 1
    Disagree = 2
    Neither = 3
    Agree = 4
    Strongly agree = 4,101,199

    but that would be silly.

    So people add the scales.

    If you wanted, you could do factor analysis on each set of 3 Likert scales (taking into account the ordinality) and then use the factor scores. These would likely be very similar to simply adding the items.

    Another possibility is to use some form of optimal scoring.

    But all that gets fairly involved.

+ Reply to Thread

           




Tags for this 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