Statistics tests of significance

[SpartanG345]

For the following tests of significance

1 tests of mean difference for paired data
- differences a calulated - forms a new random variable test its mean

2 two sample normal tests
- looks at the differences in averages

i have trouble picking which one to use for problems, i mean they both appear find the same kind of thing, how do u decide which one to

i think you use the first one if you want to proof 1 data set is different to another, without looking at the distribution of the data set

and you use a two sample normal tests i think you are testing whether two samples are from the same population?b]

 

[statdad]

The first test is used when the two "samples" are paired, or matched. Classic examples are situations in which you have pre-test and post-test data, and you want to decide whether there has been a significant change. In such cases the samples are typically correlated. In actuality, the paired t-test is simply the one-sample t procedure applied to the differences of the original data. Note: the only time the paired test even has a chance of being appropriate is when the two sample sizes are equal - the values can't be paired if the samples aren't the same size, although equal sample size alone isn't the tip you need to know this test should be used.

The second test is used when the samples are independent, and you wish to compare the two population means. The samples don't need to be the same size here. Depending on whether you are willing to assume the population variances are equal or are not equal, there are different versions.

Both test statistics use the t-distribution, and are sensitive to departures from normality. For the paired-t test, the differences of the data should be reasonably symmetric and outlier-free. For the two-sample test both samples should be symmetric and outlier free.