Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Test of difference between means

    Hi,

    I have data where I have compared the means and from visualising it I can see that one group have lower values than the other, however I am not sure how to quantify this so that I can say whether group 1 has slightly lower scores or much lower scores.

    Here is my table:
    Age Group 1 (males) Group 2 (males) Group 1 (females) Group 2 (females)
    40-44 37.3 50.3 25.8 30.7
    45-49 36.6 48.8 20.1 29.9
    50-54 35.9 47.6 18.5 28.7
    55-59 34.3 46.2 21.2 27.5
    60-64 30.3 44.6 19.2 26.5
    65-69 31.5 42.3 18.8 25.3
    70 23.6 39.1 10.5 23.5

  • #2
    You will increase your chances of useful answer by following the FAQ on asking questions – provide Stata code in code delimiters, readable Stata output, and sample data using dataex.


    If you looked in the subject index of the PDF documentation? You get it by going to help on the Stata window, PDF documentation, at the bottom index, the bottom of that subject index. The subject index includes an entry called equality test of means. To test differences in means, you would need the raw data from which you calculate whatever these things are in your table. Without estimates of the standard deviations, most standard tests won't work. Note, there is a big difference between substantive and statistical difference. The tests will test for statistical differences – the hypothesis that the means of the same. But if what you really want to talk about is slightly lower or much lower, it sounds like substantive difference in which case you have to interpret the difference relative to some understanding of the variables.

    Comment


    • #3
      Thanks. I don't have the raw data for group 2 but I do for group 1. I have standard deviations for group 2 and group 1. Will a test of means still therefore be possible?

      Comment


      • #4
        if you have the N's also, then yes - see the "immediate" form of the ttest in
        Code:
        help ttest

        Comment


        • #5
          I didn't know if I could do a ttest as I have three variables:
          Age (categorical)
          Group (binary)
          Mean score (continuous)
          Last edited by Joe Tuckles; 26 Mar 2020, 11:20.

          Comment


          • #6
            (Joe--my response preceded your follow up, but I think still is relevant.)

            Let me make common cause with Phil and expand on the substantive significance issue: I always suggest to people to look at whether the difference they found is big enough to care about in the real world. If it isn't, why bother subjecting it to a "test?" The hypothesis test only tells you if the difference is small enough that it might have arisen purely due to sampling variation. That's an entirely different question than "is the difference big enough to matter?" That's a "human being" question, not a statistical question. All this being said, in some disciplines people like to look at so-called "effect sizes" (which you can search on pretty easily) Effect sizes can be thought of as "is the difference between groups bigger than the 'natural' level of variation within the population from which your groups came." Some disciplines are very pro-effect size, and others deprecate them heavily.

            Comment


            • #7
              MALES:
              Age G1 N G1 Mean Score G1 SD G2 N G2 Mean score G2 SD
              40-44 43 37.3 9.8 880 50.3 10.3
              45-49 85 36.6 7.3 798 48.8 10.3
              50-54 85 35.9 10.3 820 47.6 10.1
              55-59 120 34.3 9.1 3743 46.2 9.8
              60-64 127 30.3 8.1 2683 44.6 9.2
              65-69 153 31.5 9.1 3947 42.3 8.6
              70+ 4 23.6 5.8 3286 39.1 8.1

              Comment

              Working...
              X