1

This is a very simple question but it has been years since I have done statistics. I am also new to this website (was recommended by a friend) so apologies if this question has been done 'wrongly'.

I am comparing several criteria between two different treatment methods of a certain disease. I have the following table as an example of results.

     Treatment A                 Treatment B
          14                          23
          8.2                         8.1
          17.4                        17.1
          0                           2

The first row is the amount of patients I have on each treatment. The next 3 rows are the average results of different criteria. Basically, how do I check if the difference in results of each criteria are statistically significant?

Currently, from searching around on the interwebs I believe I am supposed to do a T-test. I have been using this website: http://www.quantitativeskills.com/sisa/statistics/t-test.htm

I input the mean result of treatment A and B, the amount of each data, the standard deviation (calculated using excel with the =stdev() function) and with a width of C.I. set to 95%.

The website then gives me a lot of jargon which I do not understand. However, for some of my results it has given the conclusion that 'Difference not significant at 5%' and some of my results do not show that line, which signifies to me that the results are indeed significant.

Have I gone about this the right way?

user8022
  • 11
  • 1
  • 2
  • 4
    I think your question is answered in this thread: http://stats.stackexchange.com/questions/31/what-is-the-meaning-of-p-values-and-t-values-in-statistical-tests (apologies, can't seem to get links to look nice in comments) – Freya Harrison Dec 15 '11 at 11:42
  • 1
    I think it would be helpful if I copied my results from the website here. http://www.quantitativeskills.com/sisa/statistics/t-test.php?mean1=8.207&mean2=8.061&N1=14&N2=23&SD1=0.014&SD2=0.014&CI=95&Submit1=Calculate As far as I can tell, it is telling me p=1. That doesn't sound good. – user8022 Dec 15 '11 at 12:09
  • 1
    That's p=1 for the *one-sided* hypothesis that the mean of B exceeds that of A. Reading further you can see that p=0 for the *two-sided* hypothesis that the means are equal. The language is not clear, but the result is. That's because the difference of the means, 0.146, is over 30 times the standard error of that difference (0.0047). That's huge. However, since the standard deviations were both given as "0.01", which may have *zero* digits of precision, this result may be spurious: it could be an artifact of imprecision in providing the inputs. – whuber Dec 15 '11 at 16:43

1 Answers1

1

Not enough information unless you can specify some type of mean variance relationship. The t-test has no such relationship and it would be very silly to do specify one if there isn't a very good scientific justification for it, such as with count data or binary outcomes. You need to know how variable the data are. That requires some measure of the spread of data in both groups. The t-test uses that measure to create a scaled distance between mean values. If you were able to get the variance of the samples, then you can calculate the t-test statistic directly, or use it as an analytic weight in a regression analysis.

AdamO
  • 52,330
  • 5
  • 104
  • 209