This video explains some of the concepts associated with t-tests. It focuses on how to do the calculations in Excel. The difference between Excel for Windows and Excel for Mac are very, very small when using the Data Analysis Toolpak.
Note that I made a mistake in minute 8 when reporting the calculated t-value. It should be 2.11, not 2.14.
If you have not installed the Data Analysis Toolpak (which comes free with Excel), the following video will show you how to do it.
+Hehradad Rabiei If you have reason to believe that the populations (and not just the samples) have different variances, you should use "unequal variances." There are very detailed discussions of this questions in many places on the web, e.g., https://stats.stackexchange.com/questions/305/when-conducting-a-t-test-why-would-one-prefer-to-assume-or-test-for-equal-vari
+sher q Hi Shery. It sounds like you actually have 4 hypotheses (that is, 4 alternate hypotheses), all two-tails, that you want to test:
1. MaleFingerRD is different than FemaleFingerRD.
2. MaleRadialRD is different than FemaleRadialRD.
3. MaleUlnarRD is different than FemaleUlnarRD.
4. MaleProximalRD is different than FemaleProximalRD
You need to do 4 different two-sample t-tests on the appropriate data. Be careful setting up your data on Excel. Everything needs to be grouped carefully.
There are also some more complicated things you can do to prevent "alpha inflation" (i.e., avoiding false-positives), but I would guess that this would be beyond the scope of this project.
+David Dunaetz I have both alternate and null hypothesis. I'm testing for the differences between finger ridge density in male and female. (this is where it gets complicated), I also want to compare the radial, ulnar and proximal ridge density areas of the fingers,between male and female. Pleaseeeee help me im in my last 3 months of uni
+shery qures The difference between the two samples is significant if the p-value is less than .05, that is, there is less than a 5% chance of getting that big of a difference in means of two samples if, in fact, the two samples come from the same source (the same population). You choose the one-tail or two-tail p-value based on the hypothesis that you're testing.
This was a nice video but I believe that there's an error at 8:18 or before dealing with the t critical value in the last statement.
Shouldn't it be 1.76 of the onetail critical instead of the 2.14 of the two-tailed since our test was onetail?
Yes, Cassandra, I made a mistake when copying the value. The reported t should be 2.11, not 2.14. See the pinned comment by an166ful. We don't need to report the critical valued of t, just the actual value that we computed for our sample.
+aruna kumar You need to have a good reason to use the un-equal variance version of the t-test (for example, if the standard deviation of one sample was 4 times the size of the standard deviation of the other sample). But if you use the un-equal variance version of the test, you lose a lot of power. In general you should avoid it if you can. However, if you want to be very conservative and are willing to pay the price to avoid Typel 1 (alpha) error, you can always use the un-equal variance version. As a rule of thumb, most people always assume equal variances.
We have the data from a stress survey given to 8 people in Division A and the data from the same stress survey given to 8 people in Division B. The t-test allows us to use this data and make a conclusion whether the average stress level of the entire Division A is different the average stress level of the entire Division B. Stress is typically measured with instruments such as the "Perceived Stress Scale" or "Maslach Burnout Inventory." Much information on these scales can be found online.
David, thank you for you previous reply and let me ask one question. I am doing
t-test for the research. I formulated the hypothesis (like A greater
than B, one-tail) but means of t-test showed that B greater than A and
moreover the difference was not statistically significant (p>0,1).
What's the conclusion should I write ? Does it mean that the hypothesis
should be rejected? Also does it mean that the bigger T-value, the more solid the result?
In the situations you described, you made a directional (one tailed) hypothesis (A > B). However, your results came out indicating a trend in the opposite direction. So your conclusion should be that your hypothesis is not supported by this data. You need to retain the null hypothesis that there is no difference between A and B or even that B is greater than A. This means that your hypothesis still might be true, but that the data you collected provides absolutely no evidence for it being true.
I want to measure the similarity between A and B over 5 minutes using T test.
A refers to statistical Mean for 3 minutes and B refers to statistical Mean for 5 minutes for the same person.
So, what are the parameters for the T test?
two tails or one tail. tails? and Type?
if the P>0.05 does that mean A is similar to B?
To use a two sample t-test like I've shown in the video, you need to have all the data (not just the mean) for the three minute period and for the 5 minute period. You would enter the data into columns as shown on the video for each period.
If you don't have all the data, you at least need to have the standard deviation for both periods. However, you'd have to use formulas for this calculation which are not covered in this video.
Since you're getting your data from a single person, a more powerful test would be a repeated measures ANOVA. However, this is much more complicated than a 2 sample t-test. If the 2 sample t-test works, you can leave "good enough" alone.
Also, a t-test can't measure similarity. It can only tell you if two things are significantly different (i.e., that the difference between two means is unlikely to be due to chance). To understand how different the means are, you need to calculate the effect size d. I have a short video which introduces effect size which might interest you. "Introduction to effect size": https://www.youtube.com/watch?v=2AKTNvVN3Dk
Uploading contracts to an online database should not take too long, and with the right solution, there should be a way to quickly drag and drop them into folders. Of course, the contract management team may want to give some thought as to how those folders are categorized. In some industries, it may make sense to classify them by agreement type, whereas in others they may need to be grouped by timeframe or date. It is obviously important to do what makes sense for your company and to ensure everyone understands the classification system that is instituted. With this sort of well-oiled system in place, it is a lot easier to keep a handle on things.
Divide and Conquer.
This is another area that is very industry-dependent, but it is highly unlikely that any company can afford to have an entire contract team devoted to managing one portfolio. More than likely, it is more realistic to divvy up the team and the contracts so that there is a leader for each relevant sphere. The entire team will obviously have to coordinate and communicate, but resources must be allocated in the most efficient manner possible. In turn, this will allow for several individuals to keep an eye on a smaller batch of contracts, thereby facilitating those periodic reviews.
Outsource the Tedium to Technology.