Outliers live outside the inner quartile range. By statistical definition, they live 1.5 times below or above your inner quartile range.( copied from the link he ut in the descriotin box)
It depends on the data set. For example even if you had a outlier boundaries that's negative, it doesn't always apply. For examples a child birth dataset might not use any negative outliers then you would just use 0 as your range.
Thanks for this great explanation! Curious what the 1.5 factor is as it relates to the upper and lower bounds? Is this some type of outlier exclusion factor? Or just a suitable one based on the data set?
@@absentdata I found that I had to change that 1.5 to 1 for the dataset I was working with for it to pick up other values I considered outliers as well. Will do some more research on that factor and what it is exactly but I had to change it to get a better result.
I found this very helpful, thank you! I'm trying to compare different business with their respective values to find the most "different" one. Is this method ok to use in that case?
In a sense yes, you may be doing a univariate analysis of the three businesses. For example you may be looking at Revenue across all the three businesses in comparing their distributions. You can compare the shape of the distribution whether it's a normal distribution or a very skewed distribution. Skewed distributions are affected by outliers.
BRO THERE IS SOMETHING WRONG IN THIS DATA....BECAUSE WHENEVER I JUST APPLIED A DIFFERENT FORMULA TO FIND FIRST AND 3RD QUARTILE I JUST GET DIFFERENT VALUES WHICH IS NOT POSSIBLE
This is part of the accepted original formula. Essentially quartiles can be scaled to standard deviations. The math equates 1.5×IQR to under 3 standard deviations from the mean. So the 1.5 was designed in the formula to capture the majority of outliers assuming the data is normally distributed.
How this works on positive only data. I have list of numbers coming daily say. I want to know if the trends are inline. What is the best way to use that. If there is any dip or spike in the trend, how to t find out.
Well a column should be seen as a unique dimension in your data. so if you have 20,000 columns. By Excel only has 16,384. I would question the structure of your data. I would see big you can melt the columns down to rows
Welll not really because that invalidates the need for calculating the quartiles. However, I think the approach you are trying to take is more maual where you are setting the boundaries for the outliers yourself.
Yes, its a best practice to use 1.5. However, you can alter it but then release that you will lose data. It conforms to a standard normal distribution which means that it's a reliable way of saying that most of your data falls under this about 97.5%
@@absentdata No. I mean just three data. I conducted an analysis with 3 trials so I generated 3 data points. And I need to know whether I can average all those three and report it. Edited** What I meant with "analysis" is "experiment"
@@absentdata Is that the best option for my case to use to identify the outlier? Can you advise me some other stat tools I can easily apply on a calculator.
@@ellereal8216 Unfortunately I can't really prescribe what to do cause I would need to know your data and context. However I think you can just explain your logic when presenting your findings. You can also can also use percentiles.
Hello Sir. If the Result Show True. Meaning That we have to delete the data (true only) or the significance true data? (Im Using Panel Data). For eg: 2013 TRUE apple.co 2014 False apple.co 2015 TRUE apple.co 2016 False apple.co 2017 False apple.co SHOULD I dELETE ALL.. or just the true ?
Very helpful video, a quick question: Can I use this method to emit outliers in a degradation trendline? or this method is for data that fit normal distribution or some other distribution only?