Comparing Direct Measurement & Simulation Results: Paired t Test?

[Lisa!]

If we measure a quantity with 2 different tools(once by direct measurements and the other time through simulation), is it true if we compare these 2 results together by using paired t test?
Does it makes sense ?
For example we want to measure the aborbed dose of a patient, so once we measure it directly by using TLD(some kind of detector) and the other time we measure it through simulation by MCNP method. Now we want to see if these results are significantly different from each other or they're in a good agreement with each other, can we use paired t test for that purpose or is it wrong?

Thanks
PS1: Feel free to move it to the right forum and sorry if I've not posted it in the right place.
PS2: Please let me know if I've not clearly stated my question.

 

[EnumaElish]

You can use a paired t-test provided you are comparing outcomes from two random variables each of which is normally distributed, or can be transformed to normal through a weighting algorithm. In the latter case you need to perform a weighted t test. In case of non-random samples (e.g. you tested every third patient at 2pm every other day) there may be a sampling bias.

 

[statdad]

If we measure a quantity with 2 different tools(once by direct measurements and the other time through simulation), is it true if we compare these 2 results together by using paired t test?

Unless, in your two sets of data, the two items in the first row, the two in the second row, and so on, are collected under similar conditions (from same person, before and after, or something else) a paired t-test is not appropriate.

Does it makes sense ?
For example we want to measure the aborbed dose of a patient, so once we measure it directly by using TLD(some kind of detector) and the other time we measure it through simulation by MCNP method. Now we want to see if these results are significantly different from each other or they're in a good agreement with each other, can we use paired t test for that purpose or is it wrong?

This seems to meet the criterion for the paired test. Remember that a paired t-test is simply a regular t-test applied to the differences, so you should make sure the differences of the data are reasonably symmetric and have few (preferably no) outliers, as t-tests are notoriously sensitive to departures from normality.
Thanks
PS1: Feel free to move it to the right forum and sorry if I've not posted it in the right place.
PS2: Please let me know if I've not clearly stated my question.[/QUOTE]

 

[Lisa!]

Thank you, guys!
Your posts were really helpful...

Unless, in your two sets of data, the two items in the first row, the two in the second row, and so on, are collected under similar conditions (from same person, before and after, or something else) a paired t-test is not appropriate.

This seems to meet the criterion for the paired test. Remember that a paired t-test is simply a regular t-test applied to the differences, so you should make sure the differences of the data are reasonably symmetric and have few (preferably no) outliers, as t-tests are notoriously sensitive to departures from normality.

Nice explanation!

 

[blue_raver22]

I would say that it is not appropriate to use a paired t test in this situation. A paired t test is typically used to compare two sets of data that are collected from the same subjects or objects, but under different conditions or treatments. In this case, the direct measurement and simulation results are not collected from the same subject or object, but rather from two different methods of measurement.

Instead, a more appropriate statistical test to use would be a two-sample t test, which compares the means of two independent samples. This would allow you to determine if there is a significant difference between the direct measurement and simulation results. However, it is important to note that this test assumes that the two samples are normally distributed and have equal variances.

Additionally, it may be beneficial to also visually compare the two sets of data using a scatter plot or other graphical representation. This can give you a better understanding of the relationship between the two sets of data and whether they are in agreement or not.

In summary, while a paired t test may seem like a logical choice for comparing direct measurement and simulation results, it is not appropriate in this situation. A two-sample t test and visual comparison of the data would be more suitable methods for determining the agreement between the two sets of results.