Accepted Solutions
Hi @Alainmd02,
It is important to know what is the objective of your study : are you interested in the individual performances of products and/or in the comparison of products ?
In the first case, besides using visualizations and Graph Builder, I think the platform Fit Y by X could help to see the differences between initial vs. final values for each products. Since you have low sample size with non normal distributions, I would suggest to use a non-parametric method for the comparison of means (Nonparametric Test Reports). Here is how the analysis would look like on Product 1 with a Wilcoxon test :
In this case, a statistically significant difference is detected between values initial and final for Product 1. To do this analysis, you'll need your data in a stacked format (Launch the Oneway Platform), see the datatable attached containing the script for this statistical test.
In the second case, there might be several way to do it, and it will require a split data table format (one column containing initial values, one column containing final values). Here are the possible analysis :
- Doing an Test Equivalence on all products (platform Distribution): If you create a calculated difference column (Final - Initial), you can test if the difference between final and intial values are in a range of practical equivalence (0 +/- a delta you find practically equivalent).
As an example, I did it with a delta value of 1 (meaning there are no practical difference by using the products if the difference between initial and final values are in the range [-1, 1]), a target value of 0 (no difference between initial and final on this difference column), and a confidence level of 0,975 (to realize two one-sided t-test on low and high levels, each with a confidence level of 0,95). - You can also do this equivalence test by products, by using the "By" panel when launching Distribution platforms and use it for column "Product".
- You may also do a Matched Pairs Analysis (menu Analyze, Specialized Modeling, Matched Pairs) : Include your two responses (initial + final), and Product as your X group. By default JMP will launch a paired t-test, which is a parametric test, but there are other options in the red triangle to realize a non-parametric paired test. For example I used a Wilcoxon signed rank test in the script. Example here : Example Comparing Matched Pairs across Groups (jmp.com)
- You can also realize a Manova model, which will show you the differences between products and within products.
See table "Split data" to see these options.
- You may also use all initial values as a "control group" and use the final values by products as different products group to see if one or several product have a statistically significant difference compared to initial values. See table 3 and its script to see this type of analysis. Nonparametric Multiple Comparisons Reports.
As I don't know your objective or the goal behind the test or visualization, these ideas should only be seen as a (non exhaustive !) list of several possibilities with your data.
You're the domain expert so you know better what would be the most appropriate way to analyze and visualize your data based on your objective.
Hope this answer will help you,
Hi @Alainmd02,
It is important to know what is the objective of your study : are you interested in the individual performances of products and/or in the comparison of products ?
In the first case, besides using visualizations and Graph Builder, I think the platform Fit Y by X could help to see the differences between initial vs. final values for each products. Since you have low sample size with non normal distributions, I would suggest to use a non-parametric method for the comparison of means (Nonparametric Test Reports). Here is how the analysis would look like on Product 1 with a Wilcoxon test :
In this case, a statistically significant difference is detected between values initial and final for Product 1. To do this analysis, you'll need your data in a stacked format (Launch the Oneway Platform), see the datatable attached containing the script for this statistical test.
In the second case, there might be several way to do it, and it will require a split data table format (one column containing initial values, one column containing final values). Here are the possible analysis :
- Doing an Test Equivalence on all products (platform Distribution): If you create a calculated difference column (Final - Initial), you can test if the difference between final and intial values are in a range of practical equivalence (0 +/- a delta you find practically equivalent).
As an example, I did it with a delta value of 1 (meaning there are no practical difference by using the products if the difference between initial and final values are in the range [-1, 1]), a target value of 0 (no difference between initial and final on this difference column), and a confidence level of 0,975 (to realize two one-sided t-test on low and high levels, each with a confidence level of 0,95). - You can also do this equivalence test by products, by using the "By" panel when launching Distribution platforms and use it for column "Product".
- You may also do a Matched Pairs Analysis (menu Analyze, Specialized Modeling, Matched Pairs) : Include your two responses (initial + final), and Product as your X group. By default JMP will launch a paired t-test, which is a parametric test, but there are other options in the red triangle to realize a non-parametric paired test. For example I used a Wilcoxon signed rank test in the script. Example here : Example Comparing Matched Pairs across Groups (jmp.com)
- You can also realize a Manova model, which will show you the differences between products and within products.
See table "Split data" to see these options.
- You may also use all initial values as a "control group" and use the final values by products as different products group to see if one or several product have a statistically significant difference compared to initial values. See table 3 and its script to see this type of analysis. Nonparametric Multiple Comparisons Reports.
As I don't know your objective or the goal behind the test or visualization, these ideas should only be seen as a (non exhaustive !) list of several possibilities with your data.
You're the domain expert so you know better what would be the most appropriate way to analyze and visualize your data based on your objective.
Hope this answer will help you,