Reasons for weak multiple regression 1) Could you enumerate few main reasons for weak (statistically non-significant, like p > .05) results ? After running multiple regression analysis.
2) There is one construct, which plays a role of a dependent variable. However that construct composed of several attributes (measured on ordinal scale), thus there are several ordinal dependent variables. What key negative effects can arise in results if the arithmetic mean is taken for these several dependent variables, so that DV becomes one continuous value (or scale in SPSS). This transformation basically was done make life easier and run multiple regression with one scale DV and few ordinal IVs.
 A: I'm going to make a bold assumption here that you are a beginner. @Nuclear03020704 is completely right in saying that two reasons why your model did not give a significant result are that


*

*The relation is indeed insignificant.

*The model used is poorly designed, incorrect, or incapable of capturing the relationship.


There is no way to be sure of reason 1 other than by trying to remedy reason 2. If you have a complex or messy data set, and you have not learned medium to advanced techniques for building models, then you can't remedy reason 2 without learning some new stuff.
As for the ordinal scale, there is a proper model for this case called ordinal logistic regression, which is an extension of bivariate logistic regression to more levels. You can perform this test in R or other statistical software.
If you are creating a composite score from multiple ordinal attributes, then I would strongly suggest you do one or both of the following:

*

*Hit the literature hard to see how previous investigators have handled their outcome variable(s).

*Consult an experienced statistician to help you figure out a solid, supportable approach.

If your effect is real, then a poor model can easily give you a false negative. Or, perhaps worse, a "true" positive with the wrong effect. If your effect is not there, then a poor model can give you a false positive.
Refine your model.
A: Several reasons why p-values can be > 0.05:

*

*The relation is indeed insignificant.


*The model used is poorly designed, incorrect, or incapable of capturing the relationship.
Using arithmetic mean to summarize an ordinal-scale variable can be misleading. In ordinal scale, distances between any two adjacent points are not guaranteed to be the same. Mean is not the correct way to summarize an ordinal scale, and this can lead to incorrect conclusions in the regression.
In extreme case, it could really present a problem if there are only few numbers on the scale (e.g 1-3) or if the data is tending heavily on the extremes (almost all lies in lowest or highest number).
A: we generally consider p value > 0.05 as statistically non significant in a model due to following reasons:
1: p value > 0.05 is considered as probability of that feature being close to 0 in contributing the prediction for class.
Also those features which have p value>0.05 are not normally distributed.This can be checked by plotting the Norm-distr graph against those. This is also one of the consequence why p value >0.05 is considered non significant. Features having p>0.05 are generally removed from the model.
A: It is recommended. Features having p>0.05 are removed from the model because their probability of beta coefficient (which we get for each feature after regression) is considered close to zero hence keeping those features in a model does not sound significant.
For eg:
y = x+z
beta for x = 1.2 p=0.6
beta for z = 2.2 p = 0.02
here z feature is more significant then x because probability for beta of z is less than 0.02
