Statistics

[Matthew_]

Are their any keratonic statisticians out there? In fact, I won't be biased, are there ANY statisticians out there? I am in trouble with stats in my MBA project and could do with some expertise. Long shot, I know!

[GarethB]

Not exactly a statistician, but I am getting better with Minitab and statistical analysis of data at work.

Am I going to regret placing this reply?

[Matthew_]

very likely! Do you know anything about t-tests on Micro-soft excel. I can run the test and get a result but I am not sure what makes for a significant result, I am more used to linear regression. Problem is I am not sure if my sample is normally distributed or not so I have to go for a t-test. How do you take a t-test and work out if it is statistically significany or not?

[GarethB]

Ok, will do my best, not too up to speed on stats in Excel asi it is a bit basic for what I use at work.

Firstly go to tools and look for the Data Analysis option. If it is not there select Tools and then Add-ins and put a tick in Analysis tool pack. It will install the add in and prompt for the office disc to complete the install.

To determin if you have normal distribution, plot the data as a histogram/bar chart. If the general outline of the plot looks like an upturned bell than you have a normal distribution.

The null hypothesis is that there is no statistical difference. Typically a 95% confidence limits are used, which means the p-vaue is 0.05. (Don't ask why, I do not know!).

The the calculated p-value us 0.05 or less, then the null hypothesis is rejected and the data sets being compared are statistically significant.

Minitab help states

"p-value
Used in hypothesis tests to help you decide whether to reject or fail to reject a null hypothesis. The p-value is the probability of obtaining a test statistic that is at least as extreme as the actual calculated value, if the null hypothesis is true. A commonly used cut-off value for the p-value is 0.05. For example, if the calculated p-value of a test statistic is less than 0.05, you reject the null hypothesis."

There are 3 typs of t-test;

Paired two sample mean
Two sample assuming equal variance
Two sample assuming unequal variance

Paired two sample is the most common, but if you look up t-test in the Excell help, it will give the definition of each to help determine which you need.

You might like to look at ANOVA too.

I have more information at work if you need further info, but to look properly would be when I have 30 minutes lunch; usualy somewhere between 12 and 2pm!

[Matthew_]

Gareth,
That's great - that will get me started at least!
Matt

[Andrew MacLean]

Matthew

I don't have the first clue about Excel, and all I can remember from University is that the t test is a test of the relationship between two populations: the control and the treatment group.

t is expressed as a ratio of the mean socres of each population against the variability of scores in each population.

The top line is easy; you just subtract one balue from the other.

The bottom line is called the standard error of the difference. To compute it, you take the variance for each group and divide it by the number of people in that group. The variance is just the square of the standard deviation.

You add these two values and then take their square root.

I guess that all this calculation is what Excell does for you, but when I was a student in the 1070's, there was no excell

Good thing Gareth is younger than I am, and is able to deploy all this computing technology. I still have trouble with the electric kettle.



I think this is the formula for the t test.

Andrew

ps

The t-value will be positive if the first mean is larger than the second and negative if it is smaller. Once you compute the t-value you have to look it up in a table of significance to test whether the ratio is large enough to say that the difference between the groups is not likely to have been a chance finding. To test the significance, you need to set a risk level (called the alpha level). In most social research, the "rule of thumb" is to set the alpha level at .05

for more on t test, you could go to
http://www.socialresearchmethods.net/kb/stat_t.htm

My son recommends the site.

[GarethB]

Part of my stats exam at Uni was to calculate things long hand. One question and data set would take pretty much the whole exam.

I go back to getting the original DOS programes to do the stats.

The best part of a day would be spent entering the data. You would then select the analysis you want and hit go, then go home. When you got back to the computer the next day the calcualtion would be done and the printout would be jammed in the printer. There would be no way fo just rinting the data, you would have to run the whole program again!

[Matthew_]

Thanks to you both, I have got it now. I don't why Excel gets such bad press from statisticians, there seems like a bit of snobbery going on but I'm probably missing something!
Anyway it worked for me, I started playing with it, using Gareth's hints and before you knew it, I was away. Ideal because my project supervisor was away. He's back now (very helpful guy). Probably shouldn't say anymore -commercial in confidence - but I was studying a leadership development experience. The t-test helped to nail down which bits worked and which did not by proving the statistical significance of changed perceptions. I basically had to find out whether or not to reject the hypothesis that parts of the experience developed people's attitudes to leadership. Really interesting! Andrew, that web site is cracking, that's what I need statistics for social sciences not for statisticians (my brain isn't big enough for that!)

[Andrew MacLean]

I am afraid that I am not too good at using the internet as a source of information, but my son has grown up with as part of the internet generation.

I am happier with reference books, but he tells me that they are now too limiting for the information needs of the new world.

I have a feeling that this is what it must have been like to be a scholar when they invented the printing press.

Andrew

[GarethB]

The trouble I have with stats is that although changes may be statistically significant, I then have to decide with respect to method limitations are the changes of real practical benefit