GPower for Mediation Analysis and its Alternative

Поделиться
HTML-код
  • Опубликовано: 3 дек 2024

Комментарии • 54

  • @dsavkay
    @dsavkay 11 дней назад +1

    This channel is such a valuable oasis on RUclips, thank you very much for sharing these information for free!

  • @dsavkay
    @dsavkay 11 дней назад +1

    Amazingo! 💯

  • @carinahorst1295
    @carinahorst1295 4 года назад +3

    Thanks for the great video! Could you say a bit more about how to actually use GPower for JointSignificance (respectively bootstrapping with power > .80), please? I cannot follow which tests to use in the programme when you talk about it in the video. Thank you!

    • @RegorzStatistik
      @RegorzStatistik  4 года назад +8

      In this case we have two regression models. The model for the a-path has one predictor (IV), the model for the b-path and the c'-path has two predictors (IV, MED). If you have covariates in your model you add their number to that.
      I would use:
      test-family: t-test
      statistical test: linear multiple regression, fixed model, single regression coefficient
      type of power-analysis: a-priori
      Then you can decide whether to do a one-tailed or a two tailed test.
      Effect size: Here you can choose a small effect for the path (f²=0.02), a medium sized effect (0.15) or a large effect (0.35) - or anything in between. You make this decision seperately for the a-path and the b-path.
      alpha error 0.05
      Power: Here you put in the power for ONE path only, so that both powers multiplied lead to the power for the mediation you want (as I showed in the video)
      Number of predictors: 1 for the a-path (plus covariates, if you use any), 2 for the b-path.
      Then click on "calculate"
      (Technically the first regression for the a-path is not a multiple regresssion but a simple regression. But since simple regression is just a special case of multiple regression I assume that makes no difference.)

  • @nohan7580
    @nohan7580 2 года назад +1

    Great video!

  • @KasiaAdamska-j4h
    @KasiaAdamska-j4h Год назад +1

    Thanks a lot for this video, it was extremely helpful! Following from that, how would one calculate power for moderated mediation?

    • @RegorzStatistik
      @RegorzStatistik  Год назад

      Unfortunately, I don't know any tool for an easy power calculation for a moderated mediation analysis.

  • @bloooobare
    @bloooobare 4 месяца назад +1

    hi, thank you for this video! im using percentile bootstrap for my mediation analysis (PROCESS v.4), it means i can't use g power because it doesnt have a distributional assumption, am i correct? if i am, i need to base my gpower on fritz & mackinnon, right? what i'm asking is how do we choose the path effect size? is it up to the researcher or based on previous research/theoritical assumption? thank you ^^

    • @RegorzStatistik
      @RegorzStatistik  4 месяца назад

      Ideally, it is based on previous research.

    • @bloooobare
      @bloooobare 4 месяца назад

      @@RegorzStatistik okk i get it now. Thank youu

  • @pratibhashrestha8474
    @pratibhashrestha8474 3 года назад +1

    Thank you for the informative video.Does this method works for survival outcome when i have categorical exposure and mediator?

    • @RegorzStatistik
      @RegorzStatistik  3 года назад

      Unfortunately I don't have any experience with survival data.

  • @NasiMcCool
    @NasiMcCool 2 года назад +1

    Thank you for the video! How would I go about testing a moderated mediation (Hayes PROCESS model 58 for example)? Thank you in advance!

    • @RegorzStatistik
      @RegorzStatistik  2 года назад

      Unfortunately, I haven't made a video about that yet.

  • @gaara97531
    @gaara97531 2 года назад +1

    Does anything change in using the paper by Fritz if a plan do an experimental study with 4 Conditions (1 Control) and plan to do a mediation analysis?

    • @RegorzStatistik
      @RegorzStatistik  2 года назад

      I don't think that the paper by Fritz&MacKinnon is applicable here, because with your design you would have 4 IVs (4 dummy variables) for the 4+1 conditions and the paper is about one IV only.

    • @gaara97531
      @gaara97531 2 года назад

      @@RegorzStatistik thanks for the quick answer. do you have any insights on how I can approach this topic? There seems to be a lack of information on this.

    • @RegorzStatistik
      @RegorzStatistik  2 года назад

      @@gaara97531 The only thing that comes to my mind is programming a Monte Carlo simulation to calculate power (but I don't know any easy ressources how to do that).

  • @sophie8624
    @sophie8624 4 года назад +1

    Thank you for this very helpful video! You mentioned the power was .80 for the Fritz & MacKinnon table, was the significance level p

    • @RegorzStatistik
      @RegorzStatistik  4 года назад

      I am not even sure whether they have mentioned which significance level they used (they have used alpha for the a-path). So I have always assumed that they had used p

    • @hannahmcgowan152
      @hannahmcgowan152 3 года назад

      @@RegorzStatistik this was my question too. So, should we assume p < 0.05? thank you!

    • @RegorzStatistik
      @RegorzStatistik  3 года назад

      @@hannahmcgowan152 For my master's thesis I have assumed that they have based their calculations on p

    • @hannahmcgowan152
      @hannahmcgowan152 3 года назад +1

      @@RegorzStatistik Thank you for the info and taking the time to look through the article! I really appreciate it :)

  • @yensin2933
    @yensin2933 8 месяцев назад

    hi, the number of the predictor refers to the arrow pointing to DV right? If my model include arrows for direct effect and mediation effect, I have to include all arrows or only direct effect arrows as number of predictors ?

    • @RegorzStatistik
      @RegorzStatistik  8 месяцев назад +1

      All arrows poitning to the variable (from IV and MED)

    • @yensin2933
      @yensin2933 8 месяцев назад

      @@RegorzStatistik Thank you so much, is there any article that I can review for?😄

  • @ProudToBeGREEK9
    @ProudToBeGREEK9 Год назад

    what if we have more than 1 models in our study? do we need to connect them all or calculate for each one and then sum all the predictors together?

    • @RegorzStatistik
      @RegorzStatistik  Год назад

      For more than one hypothesis I would run a power analysis for each hypothesis test and then take the largest resulting sample size. Thus you would make sure that the power is high enough for all your hypotheses.

  • @jingan5625
    @jingan5625 2 года назад +1

    so helpful!!! Thank you!!

  • @katdino8935
    @katdino8935 7 месяцев назад

    Hi. If I have 1 IV and 1 mediator how many predictors shall I indicate? In my study attachment styles is my IV and resilience is my mediator.

    • @RegorzStatistik
      @RegorzStatistik  7 месяцев назад

      For the first part (predicting the mediator) there is one predictor in the model, for the second part (predicting the DV) there are two predictors (as long as the IV is continous, of course).

    • @katdino8935
      @katdino8935 7 месяцев назад

      @@RegorzStatistik I see. In that case how many predictors shall I indicate to compute for my sample size?

    • @RegorzStatistik
      @RegorzStatistik  7 месяцев назад

      @@katdino8935 I don't really understand the question given my previous answer.

  • @markuspandrea
    @markuspandrea 3 года назад

    Thank you very much for this video! Do you have any advice on doing a power analysis for a mediation analysis using the PROCESS model with three mediators? Unfortunately, I only found resources discussing analyses with one or two mediators but not three. Any help would be much appreciated!

    • @RegorzStatistik
      @RegorzStatistik  3 года назад +1

      Unfortunately not (other than running a Monte Carlo simulation).

    • @markuspandrea
      @markuspandrea 3 года назад

      @@RegorzStatistik Thank you for your quick response! I will look into that.

  • @user-vu9fy7by9p
    @user-vu9fy7by9p 3 года назад

    Hi thank you for making this video! Would you please advise which method/software can be used to calculate sample size for a multiple mediation model? Thank you in advance :)

    • @RegorzStatistik
      @RegorzStatistik  3 года назад +2

      I guess that would be a case for Monte Carlo simulation (but I haven't calculate power for a multiple mediation yet). Maybe this article can help you there:
      Schoemann, A. M., Boulton, A. J., & Short, S. D. (2017). Determining power and sample size for simple and complex mediation models. Social Psychological and Personality Science, 8(4), 379-386.

    • @user-vu9fy7by9p
      @user-vu9fy7by9p 3 года назад

      @@RegorzStatistik Thanks a lot for your help! I will check out the paper :)

  • @RathorPhD
    @RathorPhD 2 года назад

    Thank you for this video. It is very helpful.
    I am doing a study in which I am using PLS (using SmartPLS software) analysis. My research model has 3 independent variables, 2 mediating variables, and 1 dependent variable. Can I use G*power in this case to do power analysis? My sample size is 160. I need to show that power is not an issue with this sample size.

    • @RegorzStatistik
      @RegorzStatistik  2 года назад

      I don't know how the power is calculated if you use PLS.

    • @RathorPhD
      @RathorPhD 2 года назад

      @@RegorzStatistik thanks. Can you suggest some method to do it?

    • @RegorzStatistik
      @RegorzStatistik  2 года назад

      @@RathorPhD Unfortunately not, I don't know enough about PLS.

    • @mohsenyaghoubi6430
      @mohsenyaghoubi6430 2 года назад

      You can use Cohen (1992) table. See references below:
      1. Hair, J., Hult, T., Ringle, C., & Sarstedt, M. (2017). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (Second Edi). SAGE.
      2. Kock, N., & Hadaya, P. (2018). Minimum sample size estimation in PLS-SEM: The inverse square root and gamma-exponential methods. Information Systems Journal, 28(1), 227-261. doi.org/10.1111/isj.12131

  • @nono-yd1tw
    @nono-yd1tw 3 года назад

    Hi, thank you for this video. Do you know where to read about the sample size needed for a 0.80 power, for a moderated mediation with PROCESS model 14? In which there is 1 IV, 1DV, 1 mediation variable, and 1 moderator variable (dichotomous)? Thanks!

    • @RegorzStatistik
      @RegorzStatistik  3 года назад +2

      Unfortunately, I don't know any tool for sample size calculations for moderated mediation.

    • @mohsenyaghoubi6430
      @mohsenyaghoubi6430 2 года назад

      PROCESS use OLS regression similar to PLS-SEM. See references below:
      1. Hair, J., Hult, T., Ringle, C., & Sarstedt, M. (2017). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (Second Edi). SAGE.
      2. Kock, N., & Hadaya, P. (2018). Minimum sample size estimation in PLS-SEM: The inverse square root and gamma-exponential methods. Information Systems Journal, 28(1), 227-261. doi.org/10.1111/isj.12131

  • @anabenta7651
    @anabenta7651 8 месяцев назад

    What if I use 3 predictors( i’m using 3 traits of personality)? Is it wrong to do that with G power? Also, in My Thesis there’s 2 mediators and 1 criterion. Does it change anything?

    • @RegorzStatistik
      @RegorzStatistik  8 месяцев назад

      That is a much more complex model and lies outside of the scope of this tutorial.

  • @hollydonnelly2653
    @hollydonnelly2653 2 года назад

    Hi, is the effect size for mediation analysis always 0.15? I haven’t conducted my study yet so don’t know what the effect size will be?

    • @RegorzStatistik
      @RegorzStatistik  2 года назад +1

      Unfortunately not. The f2 of .15 is a medium sized effect (small would be .02, large .35).
      Getting the effect size for a power calculation is the biggest problem of this analysis. Maybe you can find possible effect sizes in the literature for the effects you are investigating. Otherwise you would have to decide how large (or better: how small) an effect you want to be able to find with your study.
      For my master's thesis I used the following argument:
      1. For my research questions there are no prior results from which I can deduce an effect size.
      2. A medium or large effect size seems unlikely because of ... (some chararcteristics of my study)
      3. A small effect would not be that relevant in practice
      Therefore, I chose an effect halfway between small and medium.

    • @hollydonnelly2653
      @hollydonnelly2653 2 года назад

      @@RegorzStatistik thank you!

  • @Bumblat
    @Bumblat Год назад

    I dont understand how to know the sample size I need with bootstrapping method

    • @RegorzStatistik
      @RegorzStatistik  Год назад +1

      Table 3 in the article by Fritz and MacKinnon. Deciding how large the a-path and the b-path should be.
      Then going to the results in the line "percentile bootstrap"