holy moly, thank you! it's true as Jeff said below...hours and hours and can't find nearly as great and informative video as this one. working on my masters thesis and needed to know EXACTLY what you just taught us. :-)
thank u so much, I am not good in English, but this explanation is much more better than videos in Russian. Ur talent of teaching ruin language border.
Dr. Verschuuren, Thanks for sharing this great video. I spent hours on Google trying to find this very thing for one of my graduate classes. Everything I read covered the principals and theory behind this, which is great, I need to know that too. But failed to cover how to make the Excel charts. Erg bedankt, Jeff
Fantastic template and excellent presentation Dr. Verschuuren! Is it possible to do the same (And Plot PIs and CIs) in Excel with quadratic, cubic or arbitrary curve-fit? I always have to go to Minitab when I need these estimates. I would like to do this totally in Excel with non-linear dependencies.
What to do ( using this excellent video) when the X values are averages from time points with different number of samples. For example: Stability batches have 8 to 10 samples at 0, 1 and 3 months, but have only 3 or 2 samples for 12 and 15 months. How to adapt the linear regression to this kind of bias? How to consider these different "weights" of each average in the X values?
Thanks Dr. Gerard … Finally I Understand the Differences in Details ..I just Wonder if you can share with us the Source Sheet in including Charts of that amazing Video ...
Nice video, but I think the definition given for a confidence interval in the first graph is a bit misleading and, to my knowledge, rather corresponds to a Bayesian "credible interval". The definition of a CI would instead be: the proportion (95 out of 100) of intervals calculated with this similar confidence level (95%) using future similar samples that will contain the true parameter (population beta). Basically, if we repeat the same sampling procedure (same population, same sample size...), then 95% of the corresponding intervals will have the true population parameter inside. Now, that being said, it is likely a matter of how "probability" is defined in the given definition in the video... If it is the "frequentist" definition (number of occurrences in future trials), that might be ok. If it is understood as the Bayesian probability (degree of belief, of certainty) then it is not ok... Hence the misleading aspect of the definition.
Dr. Verschuuren - I have the same question as "Henryluyu" below. All discussions I have seen about 95% Confidence Interval and 95% Prediction Interval are in the context of a linear relation of the form Y = aX + b. What if I have a quadratic relation Y = aX^2 + bX + c? To me philosophically, the entities "95% C.I." and "95% P.I." should exist irrespective of the relation between the dependent and independent variable. I can understand that when there are multiple variables, it can get much more challenging. For now, if I am looking at a single independent variable but a quadratic relation, I am at a loss to figure out how to calculate 95% C.I. and 95% P.I. Could you please guide me? Also, I am looking for techniques using just Excel. Thank you.
I don't know what "k" is, but for _linear_ regression the number of degrees of freedom is simply the number of data points minus two. If there are _exactly_ two data points, then the line is exactly determined, so there are zero degrees of freedom, and no confidence interval calculation is possible.
Dr. Gerard Verschuuren, Based on your experience/knowledge, would you please recommend a good reference (statistics and related) for the above mentioned determination of Confidence Intervals and Prediction Intervals if available? Thank you.
Hello sir, Thank you thank you for sharing knowledge. I wanted to fit the confidence band for the other distribution. So where I can get derivation for fitting confidence band for normal and other distributions? Please suggest any books for that ....
Dr. Verschuuren, I notice that the nature of the confidence interval line is always non-linear due to the lower and upper bound dashed lines of best fit. I was wondering if the prediction bands are always linear. On your video they look linear, however on a sample I am using they are non-linear. I am using a log-log chart could this be the reason why my prediction bands are non-linear?
Many thanks. The case shown here is about the linear fitting. I am wondering if this method is applicable to nonlinear fitting. If not, how to calculate the confidence and prediction intervals for nonlinear fitting? Best wishes.
Thank you! But what if we want calculate intervals for new X-value outside our range of data, for example, 14 in this case? Should we re-calculate X-bar, forecast new Y-value, and get the new SE, SSxx and so on; or simple use t, SE, SSxx and Xm from the present set?
If I use the formula I get with regression and get some Y from some X, what interval (prediction or confidence) to get the standard error of Y value? That is, so I can write Y±some value.
While determing the slope, intercepts,.....ssxx how can wr ifentify that the results reprrsents the same slope ,intercept... do we need to memorise the locations to identify it???
Dear Dr Gerard Thank you very much for sharing this Video but I am wondering how predication and confidence interval both can be done for a power regression curve instead of linear regression line. Will it be possible to show any power regression curve or tell us what should we change in the equations to get it done. Thank you again and looking forward to hear from you regarding this issue. BR Fahad khshim
Hello Dr. Gerard! So great videos, thank for sharing them! Just by chance... do yo know what function has to be used when I have more than 30 samples? I did a previous search and I think that the z-test has to be applied, but I don't know what excel function to use. Thanks!
