I am aware that non-standardised information is more important, but could someone shed more light on this issue? Why is that if two variables are brought to the common metric the significance disappears...
Thank you
@Witold
Here are three articles that discuss the topic you query, that may be of interest:
Repeated Measures Designs:
- Multiple Observations: http://optimalprediction.com/files/pdf/V2A17.pdf
- Single-Case Designs: http://optimalprediction.com/files/pdf/V2A25.pdf
Static Designs:
- http://optimalprediction.com/files/pdf/V2A5.pdf
Yes, non-standardized info may be more informative, but very difficult to interpret, especially when it comes to comparing variables; b/c we have no standard that makes the interpretation very easy!!
Goof Luck!!
Berhanu
Standardization always involves a loss of information, compared to the original variables.
Furthermore, the standardization tends to "flatten" the data, also to make them comparable.
Therefore, keep in mind this fundamental assumption:
1) when two or more variables are expressed in different units of measurement, it is necessary (required) standardize them;
2) when two or more variables are expressed in the same units of mesurement, it is not necessary to standardize them (indeed, standardization could be a mistake).
@Witold
Here are three articles that discuss the topic you query, that may be of interest:
Repeated Measures Designs:
- Multiple Observations: http://optimalprediction.com/files/pdf/V2A17.pdf
- Single-Case Designs: http://optimalprediction.com/files/pdf/V2A25.pdf
Static Designs:
- http://optimalprediction.com/files/pdf/V2A5.pdf
@Berhanu
There have been developments in this arena. The articles I cite above demonstrate that interpretation of comparisons of standardized variables is straightforward.
@Gioacchino:
There have been developments in this arena. The articles I cite above demonstrate that standardization eliminates "nuisance variation" in multiple subject designs, and that using raw data induces paradoxical confounding in single-case designs.
@Paul
That standardization can eliminate "nuisance variation" has long been known.
Therefore, must always be kept in mind the two basic assumptions that I have listed, before analyzing the data.
@Gioacchino
I am curious, did you look at any of the articles I linked? Which? Comments on the findings of the articles?
Your second assumption fails in two of the linked papers--in which data were assessed on the same scales, and NOT standardizing is clearly a mistake.
A major problem with legacy methods is inherently faulty assumptions. A major shortcoming of research practice is failure to evaluate one's assumptions.
Statistics is evolving, and it behooves researchers to read, understand, and evaluate with data the new methods, in order to make an informed decision--whether to cling to the past, or to travel into the future.
@Paul
All sciences and matters evolve (or devolve).
My second assumption does not fails in the linked paper.
In the first file authors analyzed data expressed in the same units, but highly variable, for which standardization can be useful (I also pointed out in my second post).
In the second file you carries out the classification of n variables, for which standardization becomes necessary.
Therefore, my basic assumptions remain valid.
I consider the matter closed.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics