Significance testing

[guyscottsouth]

I have two unequally sized samples with similar varience and broadly normal distributions. What is the best way to perform a two tailed significance test to determine if the two means are significantly difference nd how do I do this in power BI desktop?

Can a 2 tailed t test be done easily? 

[v-jiascu-msft]

Hi @guyscottsouth,

There are many build-in statistical functions in Power BI. It seems this isn't a hard work for Power BI. If you can post more details, maybe I can translate them into DAX formula, which is a functional language used by Power BI. 

Please reference these functions: https://msdn.microsoft.com/en-us/library/ee634822.aspx

and this post: significance-testing

Best Regards!

Dale

[guyscottsouth]

 I have a much larger dataset, but this is a smaller version. I want to look by question at the result score for those answering yes against those answering no, in the full dataset the variances in the two samples are similar, but the sample size is quite variable. I was hopeing to use a 2 tailed T test to test the null hypothesis that the means of the two samples are the same. Ideally i would show the differences between the means and the confidence level.

Assessment number	Question number	Answer	Overall Result
1	1	NO	24.2
1	2	YES	24.2
1	3	YES	24.2
1	4	NO	24.2
1	5	YES	24.2
2	1	YES	68.9
2	2	YES	68.9
2	3	NO	68.9
2	4	YES	68.9
2	5	YES	68.9
3	1	YES	56.1
3	2	NO	56.1
3	3	NO	56.1
3	4	YES	56.1
3	5	NO	56.1
4	1	NO	74.2
4	2	NO	74.2
4	3	NO	74.2
4	4	NO	74.2
4	5	NO	74.2
5	1	NO	66.7
5	2	YES	66.7
5	3	YES	66.7
5	4	YES	66.7
5	5	YES	66.7
6	1	NO	11.6
6	2	NO	11.6
6	3	YES	11.6
6	4	NO	11.6
6	5	YES	11.6
7	1	YES	92.6
7	2	YES	92.6
7	3	YES	92.6
7	4	YES	92.6
7	5	NO	92.6
8	1	NO	48.8
8	2	NO	48.8
8	3	NO	48.8
8	4	YES	48.8
8	5	YES	48.8
9	1	NO	55.8
9	2	NO	55.8
9	3	NO	55.8
9	4	YES	55.8
9	5	YES	55.8
10	1	NO	94.3
10	2	NO	94.3
10	3	YES	94.3
10	4	YES	94.3
10	5	YES	94.3
11	1	YES	24.1
11	2	NO	24.1
11	3	YES	24.1
11	4	NO	24.1
11	5	NO	24.1
12	1	NO	16.0
12	2	YES	16.0
12	3	YES	16.0
12	4	YES	16.0
12	5	NO	16.0
13	1	YES	76.5
13	2	NO	76.5
13	3	NO	76.5
13	4	YES	76.5
13	5	YES	76.5
14	1	NO	32.8
14	2	NO	32.8
14	3	NO	32.8
14	4	YES	32.8
14	5	NO	32.8
15	1	NO	58.3
15	2	NO	58.3
15	3	NO	58.3
15	4	YES	58.3
15	5	NO	58.3
16	1	NO	37.2
16	2	NO	37.2
16	3	NO	37.2
16	4	YES	37.2
16	5	NO	37.2

[Anonymous]

In your context, what does "overall result" indicate?

[guyscottsouth]

The overall result could be anything really, but it's probably most easily demonstrated as:

Q: Did you revise for your exam?

A: Yes

A: No

The result for each group would be the mean average score of the 'Yes' cohort and the mean average score for the 'No' ohort.

In this case, one might expect the the 'Yes' cohort average test score to be significantly different from the 'No' cohort, where as posing the question:

Q: Are you taller than 5'10"?

is somewhat less likely to show a statisticaly significant effect

I hope that helps

[Anonymous]

Thank you very much, that makes sense.

I know this is from a few years back, but did you end up using .pbix file Dale shared? We've been working to reverse engineer it a bit for our purposes but have hit a few snags.

Or, alternatively, have you created or encountered other DAX-based approaches for sig. testing in Power BI?

[guyscottsouth]

It was helpul, but I ended up going back and building something from first principles. What I would say is that the exercise is significantly easier in R or python, so I've subsequently used that

[guyscottsouth]

Apologies for the bump, but i wondered if anyone had any further advice?

[v-jiascu-msft]

Hi @guyscottsouth,

I recalled my statistics. I don't know if this could help. Maybe you review it first, then we could go further. Please check this file:https://1drv.ms/u/s!ArTqPk2pu-BkgRFQZ5z_1PDA0QpQ.

Best Regards!

Dale

[Nik-nklskrmnn]

Hi Dale, do you stil have the file? I am Looking for the exact same thing. The link is not valid anymore.

Thanks

[Anonymous]

This is awesome Dale. Are you by chance available at all to work me though how I might re-create this to generate a template table I can use on all or most of my data sets?

[guyscottsouth]

Thanks for the response, your time and effort are greatly appreciated, I'll have a look through tomorrow as you suggest and get back to you

regards

Guy