Announcement

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

  • Understanding interactions between 2 multi-categorical variables

    Dear community,

    In an experiment I conducted where i implemented different gamification strategies to see whether people would issue more reports (of problems in their environment) to the government about problems in their environment (potholes/trash/broken streetlamp, graffiti, etc). GamDes represents the strategies and the baseline is not having a strategy. I wanted to see whether personalities (referred to in the table as HEXAD_UserType) would moderate the relationship between the gamification design a user is in and the Total number of reports they would make. I utilised the following command:

    rreg Total_No_Reports ib0.GamDes ib2.HEXAD_UserType ib0.GamDes##ib2.HEXAD_UserType Age Gender Perceived_Usefulness_Before Perceived_Enjoyment_Before Willinginess_To_Recommend_Before

    and the output is:


    From the following reference: http://www.restore.ac.uk/srme/www/fa.../12/index.html I think I need to interpret the results in the following way. The GamDes coefficients represents or is interpretable as the 'boost' in total_no_reports for Personality type 2 (the baseline) residing in GamDes 1 and GamDes 2.

    So take, for example, the interaction between 2*3 from the GamDes#Hexad_UserType interactions. It represents Gamification Design #2 * Personality Type #3. The effect has to be compared to the boost of Personality type 2 in Gam Des 2, which is approximately 14.2 in the table above. Personality type #3 in GamDes #2 has a coefficient of -13.6 so being that personality type in GamDes #2 produces a lower boost/effect on the total number of reports compared to personality type #2 in GamDes #2 14.2 + (-13.6) = 0.6.

    From the reference: https://www.theanalysisfactor.com/in...ar-regression/ I could interpret the results similarly in the following way. The constant coefficient 4.2 represents the total_no_reports at the baseline (GamDes = 0, and Hexad_UserType = 2). The GamDes #1 and #2 coefficients show the 'boost' in Total_no_reports at the Hexad_UserType = 2. Thus, for example, being in GamDes 2 and being a personality (hexad_usertype) # 2 you add the coefficients together, 4.2 + 14.2 = 16.4. Now, if you are in GamDes #2 but you are personality type 3 you would have to calculate it the following way: 4.2 + 14.2 - 13.6 = 4.8 (should I also include the main effect of personality type 3, or leave it out of the calculation because it is insignificant?). Thus, being in the gamification design #2 with a personality type #3 lowers the total_no_reports made compared to being in gamification design #2 as personality type #2. The same could be said for the baseline of both. If the constant represents GamDes 0 and personality type 2 (=4.2), then being in GamDes 2 as personality type 3 (4.2+14.2-13.6) would be would be slightly higher than the constant. Or can you not compare different GamDesigns in this case?

    In any case I can denote that gamification designs is moderated at different levels of personalities in regard to the total number of reports made. Is this a correct understanding of the interaction terms?

    Kind regards,
    Michiel
    Last edited by Mike Overwater; 15 Jun 2019, 06:43.

  • #2
    Mike:
    queries like this one are at high risk of being left unreplied.
    Instead of challenging yourself to write tons of words aimed at describing what you typed and what Stata geve you back, simply copy and past code(s) and Stata outcome(s) (or share an example/excerp of your data via -dataex-) and complete your query with a brief and clear description of what is going on. Thanks.
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment

    Working...
    X