Excel - Kruskal Wallis H test
Вставка
- Опубліковано 21 кві 2017
- Instructional video on performing a Kruskal-Wallis H test using Excel.
Example file at bit.ly/3mft9bR
Companion website at PeterStatistics.com
Donation welcome at Patreon: www.patreon.com/bePatron?u=19...
Where would the World be without precious videos as this?! Thank you very much
You're very welcome. Glad it helped.
Amazing *_* thank you so much !
You are the ab-so-lu-te best.
Geen enkele tutorial legt het zo compleet, helder en snel uit als jij.
Echt, onwijs bedankt!
Bedankt. Probeer inderdaad kort maar krachtig te zijn in de videos. Toch gaan dit soort analyses veel makkelijker met SPSS of R, dus als je daar toegang toe hebt is dat zeker aan te raden. Succes met je onderzoek.
@@stikpet Bedankt! Het gaat op zich wel rap nu, denk dat ik alleen maar in de war raak van andere programma's om eerlijk te zijn ;)
Ik heb wel een vraagje als dat ok is: als de independent variable meerdere antwoorden toelaat (iemand valt in meerdere categorieen van die variabele tegelijk), kan ik deze test dan nog wel doen? Is dat op te lossen op de een of andere manier?
Lastig in te schatten, maar als je een vraag had met 'meerdere antwoorden mogelijk', dan zal je elke optie als een losse variabele moeten zien (een ja/nee variabele). Elk van deze opties kun je dan eventueel afzetten tegen een ordinale variabele met een Mann-Whitney U test (peterstatistics.com/CrashCourse/3-TwoVarUnpair/BinOrd/BinOrd-0-Introduction.html). Tevens zou je een Cochran's Q test kunnen doen om te zien of een van de antwoorden significant vaker gekozen zou zijn dan de andere (peterstatistics.com/CrashCourse/5-ThreeVarPair/binary/MultipleBinaryPaired0.html).
@@stikpet Dankjewel :)
I also thanks your video. It is very helpful and the excel file on your website is excellent too. Could you please clarify an small issue. When or why do you use a H-adj instead of the H proposed on the original paper? Thanks again. Lee
Glad you enjoyed the video. I quickly mention at around 4:30 that the adjustment, is an 'adjustment for ties'. Ties occur when a score in y occurs multiple times. The original is I think more focussed on using the test as a non-parametric alternative for the one-way ANOVA, and assumes that the y-variable is continuous and doesn't have any (or at least not many?) ties. So, use the adjusted H in case you do have ties (which often happens if you have an ordinal variable with only a few options). Hope this answers your question.
You are a god.
Thanks. Just sharing with the world my humble knowledge on these things. Glad it helped. Good luck with your research/analysis/study.
The F4 dollar addition is a valuable info
to me it was one of those little gems I picked up somewhere and now use it all the time. Glad to have inspired others now with it 🙂
@@stikpet Totally agree. Disregarded it at first, didn't catch what it was for. Then kept getting errors somewhere. Extremely grateful.
Great video! But how did you get the Kruskal Wallis function on excel? Do you have an add-in?
Glad you enjoyed the video. The video shows how to perform the KW test step by step using basic Excel functions. I did create a so-called User Defined Function (UDF) myself using Visual Basic for Applications (VBA). This I used for the orange cells in the video. These UDF's are seen as 'macro's' in Excel, so not really an add-in. If you want to use my function you can download the example file (link is in the description), and enable the macros when you open the file.
Hope this answers your question.
@@stikpet Extremely grateful for your video nevertheless. I am using it for my PhD research as we speak.
Why are there two groups (x and y) but three catagories? I don't understand where the "3" came from for number of categories.
x determines which group, y the value. So the first one is group 2 (x = 2) and scored a 1 (y=1), on row 10 for example it is group 3 (x = 3) and a score of 4. So in essence the values for x could have been 1 = brand A, 2 = brand B, etc. and y is then the score. Hope this makes sense.
Why do you need to take the absolute values of the data points when calculating the range? / Do you always need to take the absolute values?
Not sure where I mention absolute values in this video. But for calculating a range you do not take the absolute values. If the minimum temperature in a month is -5 degrees and the maximum 10 degrees, the range is the maximum minus the minimum, so 10 - -5 = 10+5 = 15.
Hope this helps.
Hi - great explanation thanks! Just to clarify, should the name of column C be r_ij to be consistent with the formulas mentioned? Thank you!
Yes, it is
is kruskal available in the data analysis option found in the data tool bar?
to my knowledge it isn't.
Is it normal to get a value like 9,59651E-07 for the sig value?
well, 'normal' is a big word, but it's not unusual. The 'E-07' means to divide the 9.59... by 10 to the power of 7 (a 1 with 7 zeros). So very small number, but that's fine. Note however that a very small p-value does not mean there is a large effect, for effect sizes you need an effect size: peterstatistics.com/CrashCourse/3-TwoVarUnpair/NomOrd/NomOrd3c.html
i want to use the H Test with ordinal data (likert type). But unfortunately i don´t understand your formular you use. For example, the formular i have works with sum of ranks. Or is your formular just another way of calculating H?
Uhm not sure but the very first step also in my video is to determine the ranks. On peterstatistics.com/CrashCourse/3-TwoVarUnpair/NomOrd/NomOrd3a.html I also show a manuel calculation in steps
@@stikpet i know the following formula (no ties) which is beeing proposed by Kruskal & Wallis, 1952 (Use of Ranks in One-Criterion Variance Analysis): H = 12/N(N+1) * Sum of R^2/n - 3(N+1). On basis of this formula i can not follow your steps, unfortunately.
@@emsif Indeed the formula you show is for 'no ties' only. Since I have ties (and with Likert scales there very comon) the video uses the formula in case there are ties.
@@stikpet with ties i see a correction term: H(corr) = H / 1 - sum of T / N^3 - 1. I just want to understand your process :) Or am i wrong?
@@emsif 'Your' H(corr) should be exactly the same result as 'my' video, just by different arrangement. The formula I use can also be found at www.leansigmacorporation.com/kruskal-wallis-test-with-minitab/ . So eventually they are the same. In the Excel file you also see another sheet 'Kruskal-Wallis Example 1' that follows 'your' formula, and as you will see gives the same result 🙂