PDA

View Full Version : responses on t-test..



horng99
02-10-1999, 10:28 AM
Dear biomch-lers:

I greatly appreciate all who responded on my request on t -tests.
Thanks again. Below is a list of responses:

1
The Student's t-test is a statistical test done on the data once it is
collected to determine if there is a statistical "significant"
difference between groups. The subject matter is irrelevant. As for
the paired and unpaired t-test that refers to the type of data groups
that you have. Eg Paired means that you have paired data like left and
right pedicle purchase from the SAME specimen, another example is if
you are testing the wear pattern on car tires of A and B on the car,
since both are tested on the same car with the SAME testing
parameters, you have paired samples...therefore a regular test is not
sufficient.

2
Two-tailed vs one-tailed t-tests refers to the normal distribution
curve of what you are interested in determining the probability of
making a Type one (alpha) error (when you reject your null hypothesis
when in fact it is true). Given by the P-Value. Whether you use a one
or two tailed depends on your testing hypothesis eg does not equal to
would be a 2-tailed and greater than or less than would be a
one-tailed test.

This is more or less of the basics-you will find all this in a
introductory undergrad statistics text.

3
t-tests (paired and unpaired) are statistical tests used to assess
significance between your variables - they are used by all disciplines
for all types of data. If you have not had any exposure to statistics
you should consult with someone who has. There are software packages
available (SPSS, SAS, BMDP) that are used to assess statistical
differences between any type of data.

t-tests are for testing for significant differences between groups. A
t-test can also be used for a single group to test for a significant
difference from zero, or any other value.

4
The correct usage of a t-test is dependent upon the data collection,
and the research question. There are basically two types: 1) paired
and 2) unpaired, which you mentioned in your post. The paired t-test
implies that the measures were taken on the same individuals on two
different occasions, or that there is some other inherent
dependency among the groups. For example, testing subjects with the
same screw would lead to a paired test. The unpaired t-test would be
used when the groups are unrelated. For example, when you speak of
testing the BMD between unicortical and bicortical screws, it would
appear that an unpaired t-test would be the most appropriate. I am
not sure what you mean by differentiating between success and
failure of unicortical groups. However, if you were testing, say, EMG
activity in unicortical screws the correct application would be a
paired t-test.


5
In a t-test it is possible to test for both one-tailed and two-tailed
significance.
If you have no prior knowledge of the type of outcome you expect, than
a
two-tailed test would be appropriate. If you have an inkling as to a
group
increasing or decreasing its value, than you may use a one-tailed test.

Every statistical software package should have each of the t-test
options
available. A very good text on Inferential Statistics is by Glass and
Hopkins.
I do not have the text in my office at the moment, so I can't give you
the
complete reference. However, it is extremely easy to read and very
intuitive.


6
I read your e-mail on the BIOMCH-L listserver about t-tests. I am not
certain what you were asking, but I think you are looking for a
software package which performs t-tests. If you have Microsoft Excel,
it can easily and quickly do the t-tests you mentioned. In Excel,
after you have entered the data, from the "Tools" menu select "Data
Analysis." If you do not see "Data Analysis" in the menu, select
"Add-ins" and check the "Analysis ToolPak" button. The "Data
Analysis" option should then appear in the "Tools" menu. In the "Data
Analysis" dialog box, scroll down to t-tests. You can select from
Paired, Unpaired assuming equal variances or Unpaired assuming
unequal variances. The last one (assuming unequal variances) is used
if you have two groups of differing sizes (unequal N). The rest
should be self-explanatory if you have used Excel before, or you can
use the Help file to determine what you need to enter.

7
All introductory stats texts describe these tests. The paired t-tests
should be used for repeated measures (before/after) or when scores are
paired for some relevant variable (say two screws into the same bone).

Concerning unpaired t-tests a test for equal and unequal variances
should be made then the appropriate formula should be used to compute
the t value.
In my opinion one should always use a two-tailed test. This tests
whether
the two measures are different or not. If the relevant variable falls
in the
correct tail then you report that it is better, stronger, faster ...
etc.

Many people use a one-tailed test but never consider what they would
do if the
results end up in the wrong tail (i.e., worse, weaker, slower ...).
If you
can honestly say that it doesn't matter than use a one-tailed test.
Most
people, however, will consider that the results are significantly
reversed and
therefore a two-tailed test is necessary.







__________________________________________________ _______
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com

---------------------------------------------------------------
To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l
---------------------------------------------------------------