For the standardized beta values or unstandardized b values, we are interested to see the level of significance. If the beta values are significant, then it means that the predictor significantly affects the outcome.

"Beta" refers to standardized coefficients while unstandardized coefficients are often referred to as "b".

Unstandardized coefficients cannot be interpreted without knowing the scale of your variables. For instance, if your variables range from 0-1, then the unstandardized coefficients are likely to be small. However, if your variables range from 0-999,999, then your coefficients will likely be very large. This is one of the reasons why we often don't interpret unstandardized coefficients unless the variables are true continuous variables and have meaning (e.g., money, weight, height). They're practically worthless for Likert scale variables because the unstandardized coefficients depend largely on the Likert scale range.

If your variables are truly continuous and have meaning, then the interpretation is entirely conceptual and context-specific.

If your variables are not truly continuous, opt to instead interpret the standardized beta coefficients. Field norms have traditionally viewed 0.20 as small, 0.5 as medium, and 0.8 as large effects but modern norms are viewing 0.05 as very small, 0.10 as small, 0.20 as medium, and 0.30 as large (see Funder & Ozer, 2019). But this is also largely up to interpretation based on your research question.

Hello Sabee,

I agree with Hohjin Im 's advice. There is no specific threshold that would universally apply for whether a standardized regression coefficient is "acceptable." Context (the variable set, degree of collinearity among the IVs, the DV, the target population, and the model's potential application) matters, and can not be ignored.

You can, of course, choose to differentiate variables for which the statistical significance test suggests the coefficient values are non-zero from those which can not be dependably distinguished from zero.

As well, the absolute value of the standardized regression coefficient may be used to determine how much relative emphasis a specific IV is given in the model, compared to the others. However, if the IVs are correlated (amongst themselves), the relative differences may be magnified compared to what the differences among zero-order correlations for each IV with the DV might be.

Good luck with your work.