Many thanks Professor Mack, a very useful lecture. One question: Often it is proposed to make use of standardised measures of dffit and dfbeta. Are we safe using the normal rules of thumb in interpreting these values and simply investigating if more than 1(5)% of the sample case values exceed 2.58(1.96)?
We don't do statistical tests on measures of influence. The table on slide 8 gives some guidelines for identifying influential data points. But when we identify a data point as being influential, this is not necessarily a problem. We just need to be aware of it. If we are worried that one or a few data points are being overly influential (that is, we fear that our conclusions might be "fragile"), we may decide to collect more data in that space to reduce the influence of any one data point.
Finally a video where the formula for Cook's Distance is explained clearly. Thank you so much!
Super useful! Good balance between mathematical and practical explanation
This is very interesting
Many thanks Professor Mack, a very useful lecture. One question: Often it is proposed to make use of standardised measures of dffit and dfbeta. Are we safe using the normal rules of thumb in interpreting these values and simply investigating if more than 1(5)% of the sample case values exceed 2.58(1.96)?
We don't do statistical tests on measures of influence. The table on slide 8 gives some guidelines for identifying influential data points. But when we identify a data point as being influential, this is not necessarily a problem. We just need to be aware of it. If we are worried that one or a few data points are being overly influential (that is, we fear that our conclusions might be "fragile"), we may decide to collect more data in that space to reduce the influence of any one data point.
thanks a lot, it was really useful