How about the p-value in the indirect effect (cognitive skills vs. word reading)? the process software only showed the confidence interval and not a p-value while the total direct effect had a p-value. Could you please explain that? thank you for the great video.
Hi Hanieh. The only way to get a p-value for the indirect effect (currently) is the Sobel test, or what Process calls the "normal-theory test. This test assumes the estimate of the indirect effect is normally distributed, but that is only true for large samples (n>400). So the Sobel test is not recommended unless the sample is large enough. Instead Process produces a bias-corrected bootstrapped sampling distribution and reports the 95% confidence interval from that. It effectively produces many (1000, 5000, etc) "new" samples (actually the samples are individuals randomly picked from the original sample so not really new data), makes an estimate of the indirect effect in each sample, and the distribution of all those estimates is used as the sampling distribution. If the confidence interval from that doesn't cross zero, then you reject the Null hypothesis and the confidence interval is reported instead of the p-value as your justification.
Hi Shamini, c' can be greater than c if there is some kind of suppression going on. Simplest one is where path a*b is the opposite direction of association compared to path c'. Example I use is student anxiety. High anxiety reduces exam performance, but at the same time, anxiety can increase time spent studying and therefore lead to improved exam performance. If high anxiety leads to increased study which then leads to better exam performance, you have a mediated association. If some anxiety is not associated with more study, then there might also be an independent direct path (path c') that leads to a decrease in exam performance. Path c is the combination of the two and is going to be weaker than either path c' or path a*b. In fact path c might be no association at all (which is why we don't use Barron & Kenny's approach any more). I hope that helps, Garry
@@power65472 thank you..the example which I have given is exactly the same as I found in my data. Then how can I predict this c is also significant. How will I write it in my result
Hi - thank you for this - the coefficient for c' was -.101, however its significance was .127 according to the table, which is higher than p = =.05, can you still consider this a valid beta?
Prof., may i know the value of sample size to bootstrap? in your video, you put 5000. but some others put 1000. on what base we can decide this value. thank you.
Hi Arnida, 5000 is fairly standard practice (you'll see it in plenty of published articles). Five thousand is now the default in Process (the first version of Process the default was 1000). I doubt that you'll get any serious improvement in using more. than 5000. Using more samples won't reduce the confidence intervals or anything, which increasing a sample size would normally do. It would only ever-so-slightly increases the precision of the estimate of where the limits should go. That's what I mean by no serious improvement. If you can access Andrew Hayes' book on Mediation and Moderation it probably has specific advice and could be cited.
Hi, I'm not really expert in G*Power - even though I'm a stats type and my name is Garry Power. I've used it a bit and find it a bit annoying (and not just that they stole my name). You can certainly find the required sample to test a single coefficient in a multiple regression - and that's probably your best bet. The indirect effect is the product of two coefficients so will need more power than for one, but it's probably as good as you'll get.
Hi Thierry, you can use either so long as you are consistent. I prefer standardised since you can compare the effects of different predictors. But if the raw units of measure make sense to your reader the unstandardised can be really useful. If one unit of change in the IV and one unit of change in the DV make sense (one year extra education is associated with how many extra dollars of income) then consider using unstandardised.