Hello Fabio,

If published studies of technical quality of the scale (in its original language) are available, you could certainly duplicate those kinds of analyses (typically including: estimates of score reliability, correlation of scores with external or criterion variable(s), and factor analysis to affirm a comparable factor structure for the translated version).

Good luck with your work!

Dear Dr Porru,

this is an interesting question. I entirely agree with Dr. Morse's answer. I would also recommend a factor analysis (principal components analysis). The following tutorial provides a step by step instruction on how to carry out a principal components analysis when using the SPSS-package:

https://statistics.laerd.com/spss-tutorials/principal-components-analysis-pca-using-spss-statistics.php

Yours sincerely,

Rainer

Hi,

consider these steps:

1) item-level analysis

- Descriptives - mean/median/mode/sd/min/max/range, etc. and difficulty/sensitivity measures if appropriate.

- Correlations matrix among the items (note: use polynomous/polyserial correlations for items with less than 5 levels, e.g., likert scales ranging from 0 to 3 or so)

- Differential item functioning (optional, there are better ways to estimate such invariance)

- Item Response Theory approach - advanced, very powerful when optimizing tests/scales.

2) scale-level analysis

- Internal consistency - you can use Cronbach's alpha if you think all the items are equally loading the factor/construct and if there is only one factor/dimension. Otherwise, use McDonald's Omega (note that for scales with with less than 5 levels, you should use ordinal Alpha or Omega)

- Optionally you can use other indices such as Revelle's Beta (indicating homogenity), Cronbach-Mesbah Curve (indicating dimensionality), but there are better means for this...

3) Model level analysis

- if you have no a priori model (there is no hypothesis how the items should group together), then you can use exploratory factor analysis (data reduction, principal compontents, % explained variance, commonalities, rotations etc).

- if you have a priori model structure, you can use confirmatory factor analysis. Which is a very powerful tool - verify the hypothesis of the data structure (covariance matrix), estimate fit, reliability, latent scores, dimensionality, loadings, parsiomony, measurement invariance, etc.

Ideally, it would be the best if you have the results of previous research on the original scale so you can compare your results. As for the validity, it would be inevitable to demonstrate that your scores/results (item-level, scale-level, and/or model-level) approximate the original parameters and that they correlate with similar (other) constructs (as the original scale/test).

Good luck!

M.