Can you please tell me when to use Cramer's, when to use chi-squared and when Cressie..? I have a huge data, percentage of expected count less than 5 is greater than 25% and in some cases, data has more than 2500 expected count. I can segregate the data based on expected count but I don't know which test to use for goodness of fit. Edit: By segregate I mean I can calculate the chi square or any other test separately for data having low expected counts and data with high E.C.
Since you mention 'goodness-of-fit' I assume you are indeed performing a 'goodness-of-fit' (GoF) test. The Pearson chi-square GoF test becomes unreliable if any of the cells have an expected count less than 5 (Peck & Devore, 2012, p. 593). The first thing to wonder is if indeed you have nominal variable and not ordinal. If you have an ordinal variable another test might be more suitable. If it is indeed a nominal variable and you simply have a lot of categories that have only a few counts, one option used is sometimes to group these into an 'other' category. If all of that doesn't fit your needs a so-called exact multinomial test can be used, but haven't figured out how to do that yet with Excel. Than the effect size. If you want to use Cramer's V, it is slightly different for a GoF test. Have a look at: ruclips.net/video/1Oe50im0XGU/видео.html . Cramer's V is an effect size measure, which is something else than a statistical test. Perhaps have also a look at peterstatistics.com/CrashCourse/2-SingleVar/Nominal/Nominal-2a-Test.html Hope this helped.
@@stikpet Wow so quick response and it helped a lot! I just have one more question if you don't mind. In a 2 lane road (one direction), many types of vehicles are present and are not equally distributed in both lanes. Therefore to estimate one vehicle type say two-wheelers, being behind/following say car depends on composition of both these vehicle types in that particular lane. How can I consider the composition of both vehicle types? If there was only one lane, probability of cars being leaders for 2W would simply be number of cars divided by total number of vehicles. To do this I multiplied the percentages of 2Ws and Cars but summation won't give 100%, so I extrapolated each probability to 100%. If it's incorrect, is there a proper method developed to do this analysis?
For example, 30% of total number of cars are in lane 2, 10% of total number of 2Ws are in lane 2. Multiplication would give 3% of total vehicles to be in lane 2, similarly around 6% are in lane 1. Total of lane 1+lane 2 = 9% which should have been 100%. So I am extrapolating 3% and 6% to 100%. I strongly feel this is wrong but I don't know any better.
Imagine a cross table Type | Lane 1 | Lane 2 | Total Car| | | 2Ws| | | Total| | | In your calculation you have the percentage based on the row totals. 30% of the total number of cars is in lane 2, and 10% of the 2Ws are in lane 2. What you need/want are the percentages based on the column total.
@@stikpet OMG bro I feel overwhelmed from your replies.. From that I would get to see if there are lane preferences from Chi-squared. I want to see if a vehicle has preference to "who he follows" Like I observed that no vehicle types like to follow 2Ws. 2Ws are usually following themselves. I have a huge data but not for trucks, buses, light commercial vehicles. I'm going to merge these as "others" but still I'm not clear of the procedure. I'll give the example of actual data I have. These are the observed number of vehicles 3Ws are following. So from this table can I tell if 3Ws does not normally follow Cars, they more often follow 3Ws etc.? I have this table for all vehicle types. If I could predict the probability of 3Ws entering in a lane after a car, that would be next level but I must prove this first. (3Ws are following 11 cars in lane 1) No. of Cars present in field >> 2Ws > 3Ws Type | lane 1 | lane 2 | Total Car | 11 | 6 | 17 2W | 14 | 16 | 30 3W | 19 | 22 | 41 oth | 5 | 4 | 9 Tot | 49 | 48 | 97 Thank you Surya
Thank you, this video was so helpful. Do you know, If I wanted to use FDR instead of Bonferroni correction, is there a similar way to adjust the significance level using FDR in excel?
Glad it was helpful. I haven't looked into FDR myself, but a quick Google search did pop-up this video: ruclips.net/video/S-F_R_WKNfQ/видео.html hope that helps.
short and on point, nice!
Thanks. Exactly what I'm aiming for in my videos :-) Good luck with your analysis.
you are the best bro! thanks!
You're welcome! Glad it helped.
quick and simple, thank you!
Glad it helped!
you saved meeeeeee THANK YOU
glad the video helped.
Can you please tell me when to use Cramer's, when to use chi-squared and when Cressie..? I have a huge data, percentage of expected count less than 5 is greater than 25% and in some cases, data has more than 2500 expected count. I can segregate the data based on expected count but I don't know which test to use for goodness of fit.
Edit: By segregate I mean I can calculate the chi square or any other test separately for data having low expected counts and data with high E.C.
Since you mention 'goodness-of-fit' I assume you are indeed performing a 'goodness-of-fit' (GoF) test. The Pearson chi-square GoF test becomes unreliable if any of the cells have an expected count less than 5 (Peck & Devore, 2012, p. 593).
The first thing to wonder is if indeed you have nominal variable and not ordinal. If you have an ordinal variable another test might be more suitable. If it is indeed a nominal variable and you simply have a lot of categories that have only a few counts, one option used is sometimes to group these into an 'other' category. If all of that doesn't fit your needs a so-called exact multinomial test can be used, but haven't figured out how to do that yet with Excel.
Than the effect size. If you want to use Cramer's V, it is slightly different for a GoF test. Have a look at: ruclips.net/video/1Oe50im0XGU/видео.html . Cramer's V is an effect size measure, which is something else than a statistical test.
Perhaps have also a look at peterstatistics.com/CrashCourse/2-SingleVar/Nominal/Nominal-2a-Test.html
Hope this helped.
@@stikpet Wow so quick response and it helped a lot! I just have one more question if you don't mind. In a 2 lane road (one direction), many types of vehicles are present and are not equally distributed in both lanes. Therefore to estimate one vehicle type say two-wheelers, being behind/following say car depends on composition of both these vehicle types in that particular lane. How can I consider the composition of both vehicle types? If there was only one lane, probability of cars being leaders for 2W would simply be number of cars divided by total number of vehicles. To do this I multiplied the percentages of 2Ws and Cars but summation won't give 100%, so I extrapolated each probability to 100%. If it's incorrect, is there a proper method developed to do this analysis?
For example, 30% of total number of cars are in lane 2, 10% of total number of 2Ws are in lane 2. Multiplication would give 3% of total vehicles to be in lane 2, similarly around 6% are in lane 1. Total of lane 1+lane 2 = 9% which should have been 100%. So I am extrapolating 3% and 6% to 100%. I strongly feel this is wrong but I don't know any better.
Imagine a cross table
Type | Lane 1 | Lane 2 | Total
Car| | |
2Ws| | |
Total| | |
In your calculation you have the percentage based on the row totals. 30% of the total number of cars is in lane 2, and 10% of the 2Ws are in lane 2. What you need/want are the percentages based on the column total.
@@stikpet OMG bro I feel overwhelmed from your replies..
From that I would get to see if there are lane preferences from Chi-squared. I want to see if a vehicle has preference to "who he follows" Like I observed that no vehicle types like to follow 2Ws. 2Ws are usually following themselves. I have a huge data but not for trucks, buses, light commercial vehicles. I'm going to merge these as "others" but still I'm not clear of the procedure. I'll give the example of actual data I have.
These are the observed number of vehicles 3Ws are following. So from this table can I tell if 3Ws does not normally follow Cars, they more often follow 3Ws etc.? I have this table for all vehicle types. If I could predict the probability of 3Ws entering in a lane after a car, that would be next level but I must prove this first.
(3Ws are following 11 cars in lane 1)
No. of Cars present in field >> 2Ws > 3Ws
Type | lane 1 | lane 2 | Total
Car | 11 | 6 | 17
2W | 14 | 16 | 30
3W | 19 | 22 | 41
oth | 5 | 4 | 9
Tot | 49 | 48 | 97
Thank you
Surya
Thank you, this video was so helpful. Do you know, If I wanted to use FDR instead of Bonferroni correction, is there a similar way to adjust the significance level using FDR in excel?
Glad it was helpful. I haven't looked into FDR myself, but a quick Google search did pop-up this video: ruclips.net/video/S-F_R_WKNfQ/видео.html hope that helps.
Hello Sir, Can we calculate Cramers'V if the no of rows and no of columns are same. because in your video, you take min between row and Cols
Yes, no problem. If for example nr. of rows = nr. of columns = 4, then q = min(4,4) = 4.
hello sir TY for this video but can you tell me how did you calculate the person chi square value x^2 ?
no problem, I've a separate video on that: ruclips.net/video/dgjHsv8FBYU/видео.html
i think i may love you