Bootstrap Confidence Interval with R | R Video Tutorial 4.5 | MarinStatsLectures

[@marinstatlectures]

🦄 In this R video, we learn to generate a confidence interval using a bootstrap approach, step by step without using an R Package. #Bootstrapping in statistics is a resampling based approach useful for estimating the sampling distribution and standard error of an estimate. The R script accompanying this video contains all the R codes created in this video and more, so make sure to download and explore it further (statslectures.com/r-scripts-datasets) Want to support us⁉️ You can Donate (bit.ly/2CWxnP2), Share Our Videos, Leave Comments and give us a Like 👍🏼! Either way we thank you 😊

[@elenaguzman9413]

When I read this concept in my textbook, I had no clue what they were saying. Thank you so much for explaining this so well. I can finally do my homework :)

[@yangyang6008]

Hi there, thank you for the great tutorial! In the "Permutation Hypothesis Test" video, you resampled (without replacement) from the entire sample, however here you are resampling (with replacement) from particular samples (i.e., casein or meatmeal). Could you explain the difference?

[@yt-1161]

where can we learn these the good sides and drawbacks of these methods or the meaning of those tests (wilcox test, etc) how to interprete these ? Which source did you use ?

[@khiathompson33]

Thank you for another helpful, easy-to-follow tutorial! :D

[@marinstatlectures]

You’re welcome :)

[@omphileashley3142]

excellent content.

[@ChenuliSooriyaarachchi]

Thank you so much ur engaging and useful videos

[@MADDOG3333777]

Really appreciate your videos! Informative and you do an amazing job of explaining step-by-step.

[@toko2519]

@6:30 what if instead of vertical resampling, I wanted to resample rows ? So like If I have one column that is Time and 2nd column is the data points for those times. Then i want to resample just the data for that time ? How would that be done ?

[@jeffreylin235]

I am just curious whether the reviewers of the manuscript will have some issues with using a bootstrap approach in the analysis when the sample size is not large. The classic approach is non-parametric methods.

[@marinstatlectures]

id say they shouldn't. non parametric approaches work for hypothesis testing, but dont lend themselves very well to point/interval estimation (such as confidence intervals, etc)...they are also quite limited in the sorts of estimates they can deal with, while a bootstrap approach opens things up to much more interesting and novel estimates.

[@lauramaass8098]

Awesome videos! Is what you say at 10:12 relevant for every scenario? Essentially it is saying (I believe..) that if just 2.5% of our differences in bootstrap means changes from positive to negative (or vice versa, depending on the sample) we accept the null that there is no difference between the populations." I'm not sure if I'm saying that right.

[@marinstatlectures]

yes, sort of. for any confidence interval (not just one built using a bootstrap approach), if the 95% confidence interval contains the "null value" (in this case 0...the difference in means/medians is 0)...that would be the equivalent of having a p-value larger than 5%. so, here since the 95% confidence interval contains 0...we are willing to accept that the difference in means is 0...that there is no difference. this means that for H0: diff in means=0 vs. Ha: diff in means not= 0, we would fail to reject the null...we dont have enough evidence to rule the null out (you've stated this as "accept the null"....it is better to state it as "fail to reject" as the phrasing "accept there null" sort of implies that we accept its truth...and really, we never "prove" a null....we assume it to be true, and then try to provide evidence against it...so we have failed to reject it.) an important related note is also discussed in the video, comparing "statistical significance" to "scientific significance". here the confidence interval contains 0 (and the p-value would be larger than 5%)... so we fail to reject the null....BUT...while not "statistically significant", it does appear that the feed types may differ a bit, and the lack of statistical significance may be due to a small sample size (and low power to detect a difference). we really want to use "statistical significance" and "p-values" as a sort of guide, but not as a "magic number". i hope that helps clarify things.

[@zainabkhan2475]

I subscribed your channel just after hearing your kids voice!

[@pynck1000]

Such amazing content. I love it, it helps a lot, thanks.

[@elenaguzman9413]

In your example, the data has 2 levels. Does this mean, that the data needs to be converted to factors to conduct bootstrap confidence interval in R?

[@mehdifirouznia6011]

Thanks for very helpful series. My question is, in the last video of this series, you create the "BootstrapSamples" by sampling from the "variable" which contains both factors "casein" and "meatmeal", but here sampling was done separately for each group. Would you explain what is the difference.

[@marinstatlectures]

Sure, when testing a hypothesis that there is no difference in groups, we sample from both groups at the same time, as our null is that they aren’t different, and so we start with that assumption. For a confidence interval, our focal point in the sample estimate...the observed differences, and so we start from there. We allow the groups to be different, and we see how the estimate varies when allowing the groups to be different. Hope that came across clearly

[@azmal6158]

How would I do this but lets say I have data for three habitats across two years and I wanted to weight one of the areas in the resample?

[@phillipawodutire7114]

Hello. I need your help. I already estimated the bootstrap parameter estimates. I want to obtain the BCa CI but I couldn't get the code online. Can someone help me out

[@rutholiveirarase]

Congrats ! This video is perfect ! You helped me a lot! Thank you so much!

[@marinstatlectures]

great to hear! youre welcome :)

[@justinatuulikefonangolo5509]

Thank you very much for your informative videos. I have a question on how to apply bootstrapping to a sample with two categorical variables? My sample for pH is taken at two different seasons and 3 sampling locations, which resulted in having 3 columns: pH, Season and Location. How do I get a pH bootstrap resample for each season and location?

[@marinstatlectures]

hi, it's difficult to answer, as it depends on what it is you want to do exactly. are you testing some hypothesis? or calculating a confidence interval for some estimate? the way you generate resamples will depend on what it is you are trying to do...

[@justinatuulikefonangolo5509]

@@marinstatlectures yes I am testing hypothesis but my samples are too few to apply ANOVA, therefore I want to create replicates before applying the ANOVA . I want to test if there is any statistical difference between pH taken during the wet season and pH taken during the dry season.

[@DimitrisAndreou]

Would it be possible to apply this approach if your original observations had weights associated with them? And instead of estimating the average, it would be the weighted average.

[@marinstatlectures]

yes, but you would have to make sure to account for the weighting. you could either resample, but make sure that the probability of an observation being selected (in the resampling) is proportional to its weight, or you could take resamples, and then take a weighted average from there using the weights to adjust. the main point is yes, you can do this for weighted samples, but you have to make sure that you are accounting for the weights appropriately in the resampling process.

[@DimitrisAndreou]

@@marinstatlectures thanks for the reply! The re-weighting business was making my life too complicated (and couldn't trust the confidence intervals), so I ended up just expanding each observation to N observations, based on its weight (they happen to be integers), hence avoid working with weights at all. Thanks again!