I am testing if kit 1 is better than kit 2. I have seven samples, each with two slides, one slide went into each kit. Is the data paired or unpaired?
Bonus, what statistical test would I used to compare kit 1 vs kit 2 (ie is there a statistical diffrence in yeild), but also if the results from them corelate ie if for a sample both kits did badly, or if one kit did better.
Regarding normal distrobution, normality tests are saying that one kit is normally distrobuted, but the other isnt.
Thanks!!
edit, I've done some histograms ad they are all lent over to the leftside
Your data is paired because each of your seven samples has pairs of results (i.e., kit 1 yield and kit 2 yield). Since your data is paired, you can run a correlation and this will tell you the extent to which the yield from each kit is consistent (i.e., does greater yield from kit 1 predict greater yield from kit 2). To test your hypothesis that the yield of kit 1 and kit 2 differ, you would conduct a paired-sample t-test (see link below to do online). T-tests do assume normality and this means if you plot a histogram of your kit 1 yields and another histogram of your kit 2 yields, these should look like bell curves. But the results are ‘robust to violations of normality’ so unless your histograms are way off, it’s not that big of a concern. Hope this helps some, Rebecca. ~ Kevin
https://www.graphpad.com/quickcalcs/ttest1.cfm
(group one and two become your kits; choose “paired t test”)
I also suggest to apply paired t test. If the test results in the significance of he difference between kit 1 & kit 2 then the kit having greater mean can be regarded as better. On the other hand, if the difference is found to be insignificant then the two kits can be regarded as homogeneous with respect to their effects.
In order to know whether they are correlated, I suggest to compute Kearl Pearson's correlation coefficient and test the significance of the calculated correlation coefficient either by t test or by F test. If the the calculated correlation coefficient is found to be insignificant then the two kits can not be regarded as correlated.
If you have non-normal distribution you can use Non-parametric tests. For exampel Spearman Rank or Kendall tau for correlation, Mann-Whitney U Test for comparing two independent groups, Wilcoxons Mathced pairs test for two dependent groups (paired data).
If you mean that each of your 7 samples was split in two slides and enter only one condition (kit 1 or kit 2) you have unpaired data. Else, paired.
From your explanation, I have understood that the data you have are not paired data.
The data appear to be paired and this is the safer assumption. If the actual pairs are uncorrelated then the paired t test and and independent t test are essentially equivalent (i.e., when the correlation between samples = 0).
I am still not clear on how are the data paired as mentioned by Thom S Baguley.
More clarification is needed.
I think there may be some confusion because of the most common ways independent-sample and paired-sample t-tests are conducted and how they’re taught in stats classes. Usually we automatically know which kind of test to use because independent samples t-tests have an obvious “unit of analysis.” The unit of analysis is a thing we’re studying, like a person (in psychology) or slide (in biology) or crop (in agriculture) or so forth. The most common examples of paired-sample t-tests use time. For example, we might measure a person’s knowledge before and after an educational intervention or we might be interested in how the size of a culture grows over time. Let’s step back and consider the unit of analysis in each example. Our unit of analysis is a pair of values from a person at two times or the size of biological cultures at two times. But there is no requirement that the pair that is our unit of analysis include time. For example, we might measure something in husbands and wives, who form couples that we can use as our unit of analysis. We might pair parents and children into a parent-child dyads as our unit of analysis. One way you can tell you have a meaningful pair is if you can correlate the measures, and we can correlate qualities of parents and children, or husbands and wives. For example, do husbands and wives share moral values? Even if we have two “things” we’re taking measurements from, if they are meaningfully paired together, then the pair is the unit of analysis. In Rebecca’s case, she has samples and a sample is her unit of analysis. She is taking two measurements of each sample, with kit 1 and kit 2. Those two measurements are meaningfully paired because they come from the same sample. She can correlate the two meaningfully to test if greater yield from one kit predicts greater yield in the other. And to test her hypothesis that the yields of kit 1 and kit 2 differ, she could conduct a paired-sample t-test. Hope this helps, Dhritikesh & Everyone! ~ Kevin
A good explanatory answer has been given by Kevin Grobman.
Thanks to him.
it looks like data is not a paired one . Since you are using 2 different kits to compare the outcome it can be considered as comparing two independent samples. you can use either independent t test or Mann Whitney U test for significance.
Ravishankar Suryanarayana is correct. I have also observed that the data are not paired.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics