Very useful. But if you are changing variables, are you not assuming there is an effect and changing the data to fit that? I understand we are looking for trends in the data, and influential outliers should be dealt with on a case by case basis. But when the data is more ambiguous and we are close to having significant results, how do we know when to "tip the scales"? I find myself in this situation....... I collected data from 68 participants and my data is close to significant...... After conducting assumption checks and removing outliers it is just about significant (from p=.090 to p=.047). Also, when the adjusted R. squared is very low, how do we interoperate this? Im in a bit of a dilemma, on the one hand I could claim my model is (just about) significant, predicts 10% of variants and has B = .5 , but on the other hand, it only just about hits the mark. I don't want to claim an effect if there isn't one, and if by removing data it changes the result from insignificant to significant, it feels extra wrong. So, in summary....... As long as I can claim my adjustments are warranted, there isn't an issue right?
This has helped me so much with my analysis, thank you so so much!!!
Very useful. But if you are changing variables, are you not assuming there is an effect and changing the data to fit that? I understand we are looking for trends in the data, and influential outliers should be dealt with on a case by case basis. But when the data is more ambiguous and we are close to having significant results, how do we know when to "tip the scales"? I find myself in this situation....... I collected data from 68 participants and my data is close to significant...... After conducting assumption checks and removing outliers it is just about significant (from p=.090 to p=.047). Also, when the adjusted R. squared is very low, how do we interoperate this? Im in a bit of a dilemma, on the one hand I could claim my model is (just about) significant, predicts 10% of variants and has B = .5 , but on the other hand, it only just about hits the mark. I don't want to claim an effect if there isn't one, and if by removing data it changes the result from insignificant to significant, it feels extra wrong. So, in summary....... As long as I can claim my adjustments are warranted, there isn't an issue right?