"Statistics Help Needed" Topic

I need some statistics help.

Suppose that I had a number of different companies, and had calculated the percentage increase in sales from 1997 to 2007 for each. For example:

1: 277%
2: 168%
3: 347%
4: 190%
5: 176%

and so on.

What I want to do is to compare them to test the theory that one has increased sales at a rate that is statistically significantly larger.

I'm just stymied on this one. It's been too long since college statistics.

Sounds like you need standard deviation, but it won't tell you anything much unless you have a lot more figures; with 5 sets of figures it's worthless, with 50 it might tell you something.

I thought that was the case with standard deviation (I can do that in excel), but didn't' know if there was anything else that might work with a smaller n. I actually have eight figures, but know that's not enough.

Does the US government publish stats on sales growth for comparable sectors to the 8 companies you are working with? That would give you a larger baseline.

Now, will that let you use a true test of statistical significance? I do not know.

No – 8 figures isn't enough to give a statistically meaningful measure; you won't get anything more (or less) convincing than a "golly, that seems a bit high" gut call has already provided, unfortunately.

Dom, who must learn to type….

The question is not clear to me, therefore you must be eliminated. Anyway, do you want to know if the companies as a group have significantly increased sales? If so, you can build a confidence interval (95% is usual) and see if it includes the value zero.

In your case..
Mean 232
Std Dev. 77.9
S.E. of the Mean 34.8
Lower 95% Conf Limit for Mean 135
Upper 95% Conf Limit for Mean 328
Geometric Mean 222
S.E. of Geometric Mean 31.6
Lower 95% Conf Limit for Geo. Mean 150
Upper 95% Conf Limit for Geo. Mean 330
Range 179
Sample Size (n) 5

So yes, those five companies, on average had an increase in sales significantly greater than 135%, 150% if you take into account the data are not normally distributed and analyze the log of the data :o)

No applause please, save it for the end.

Streitax is laying down some quality egghead. However, without knowing the number of sales, we can't know if the percentages are comparable.

Also, since you are comparing data points that are separated by a decade, you might also want to consider market share. If you don't know the total number of sales, a proxy may be the sales the company earned out of the total of all five companies.

How many sales for each company?

I think you may be looking for 't-test,' though it might not apply in this case:

link