Hi! I found some research that looks at the same relationship that I am going to look at (2 possible groups on the independent variable and 6 possible groups on the dependent variable). In this article I found the effect sizes (r) were all between .15 and .30. Can I use those numbers in the window at 0:51?
Hello, when you say, "the effect sizes (r) were all between .15 and .30," do you mean the expected percentage difference is between 15% and 30%? You wrote "r" and I'm wondering if that means a correlation. If r, how would you convert that to a set of rates that can go into the two cells? There are six possible groups in your dependent variable. This G*power set up is for a binary variable (two possible values). Curious what you think about that.
You have to decide that based on a theory, data, some kind of educated guess, a collective decision, etc. Let's use an example of high school graduation rate. In this state, the average graduation rate is 80%. Talking with people who developed a special program to encourage high school graduation, you decided that the intervention program will improve this rate to 85%. So you want to detect the difference of 5% with statistical significance (and decides on a sample size that achieves that). You will plug in 0.80 and 0.85. Another situation. You are proposing a survey data collection to your client. The survey's most important question is "Do you trust your community leader?" and responses are Yes or No. You have no idea how people will respond, but somehow you feel like you want to detect the difference of 10% (between the YES group and the No group) with statistical confidence. With the high school graduation example, you knew you wanted to set one group's % to 80% based on the actual data. In this case, you have no idea. So I would set it to .45 and .55 (like the example in the video) because that (rates around 50%) would give you the most conservative power estimate (the midpoint between 0% and 100% contains the largest uncertainty, so the power analysis result is conservative; you are not overestimating things).
I only know this one www.psychologie.hhu.de/fileadmin/redaktion/Fakultaeten/Mathematisch-Naturwissenschaftliche_Fakultaet/Psychologie/AAP/gpower/GPower31-BRM-Paper.pdf
Hey, thanks for your video. But I've got a question. I have a within-subject-design, how do I do that now. The sample size doesn't include the information about how often a participant goes through all of these conditions. Can oyu help me?
Hi! How does this work if we have multiple predictor variables? Would you just input the alpha you get once Bonferonni corrected or something like that?
Hmm, I am not sure. I imagine the number of predictor variables is usually small (like 5) and I think it won't affect statistical power much. The type of adjustment you are talking about is something I tried in the past when I have multiple hypotheses.
すごい助かりました！You don't have other videos on statistics I guess? But you have a variety of content in the channel. 自分も色々趣味もってるつもりの人ですので、いいチャンエルと思って挨拶言おうと思いました。外国人でだから漢字があってるかはわからないんですけど、‘‘感が合う‘‘人だと思います。よろしければ、linkedin とかそういうのあったら連絡取りたいです。やば！もう一つのチャンエル今見た、有名人だった😂 ま、挨拶は挨拶で大丈夫かな・・・
Hi! I found some research that looks at the same relationship that I am going to look at (2 possible groups on the independent variable and 6 possible groups on the dependent variable). In this article I found the effect sizes (r) were all between .15 and .30. Can I use those numbers in the window at 0:51?
Hello, when you say, "the effect sizes (r) were all between .15 and .30," do you mean the expected percentage difference is between 15% and 30%? You wrote "r" and I'm wondering if that means a correlation. If r, how would you convert that to a set of rates that can go into the two cells?
There are six possible groups in your dependent variable. This G*power set up is for a binary variable (two possible values). Curious what you think about that.
Hi! Can I do the same for ordinal logistic regression aka a proportional odds model?
Hi! How to get the numbers you need at 0:51? Where can I get them from SPSS output or how to calculate them... Thank you
You have to decide that based on a theory, data, some kind of educated guess, a collective decision, etc. Let's use an example of high school graduation rate. In this state, the average graduation rate is 80%. Talking with people who developed a special program to encourage high school graduation, you decided that the intervention program will improve this rate to 85%. So you want to detect the difference of 5% with statistical significance (and decides on a sample size that achieves that). You will plug in 0.80 and 0.85. Another situation. You are proposing a survey data collection to your client. The survey's most important question is "Do you trust your community leader?" and responses are Yes or No. You have no idea how people will respond, but somehow you feel like you want to detect the difference of 10% (between the YES group and the No group) with statistical confidence. With the high school graduation example, you knew you wanted to set one group's % to 80% based on the actual data. In this case, you have no idea. So I would set it to .45 and .55 (like the example in the video) because that (rates around 50%) would give you the most conservative power estimate (the midpoint between 0% and 100% contains the largest uncertainty, so the power analysis result is conservative; you are not overestimating things).
Kazuaki Uekawa thank you for the answer
@@MaLifeIsARevolver You're welcome!
thanks. do you have any more sample documents for logistic regression with g power. can you share if you have
I only know this one www.psychologie.hhu.de/fileadmin/redaktion/Fakultaeten/Mathematisch-Naturwissenschaftliche_Fakultaet/Psychologie/AAP/gpower/GPower31-BRM-Paper.pdf
Hey, thanks for your video. But I've got a question. I have a within-subject-design, how do I do that now. The sample size doesn't include the information about how often a participant goes through all of these conditions. Can oyu help me?
Sorry I didn't check the message for a while. I am not sure what you mean. Can you say more?
Hi! How does this work if we have multiple predictor variables? Would you just input the alpha you get once Bonferonni corrected or something like that?
Hmm, I am not sure. I imagine the number of predictor variables is usually small (like 5) and I think it won't affect statistical power much. The type of adjustment you are talking about is something I tried in the past when I have multiple hypotheses.
すごい助かりました！You don't have other videos on statistics I guess? But you have a variety of content in the channel. 自分も色々趣味もってるつもりの人ですので、いいチャンエルと思って挨拶言おうと思いました。外国人でだから漢字があってるかはわからないんですけど、‘‘感が合う‘‘人だと思います。よろしければ、linkedin とかそういうのあったら連絡取りたいです。やば！もう一つのチャンエル今見た、有名人だった😂 ま、挨拶は挨拶で大丈夫かな・・・
Glad this video helped you! I have a linked in account!