Hello Rish,

1. Spearman correlations are fine for purely ordinal measures or a mix of ordinal/interval measures. (A Spearman correlation is nothing but a Pearson correlation run on ranked values anyway.) The interpretation is largely the same except that it quantifies the degree of linearity in ranked scores (rather than the original scores).

Correlate all or just some scores? That depends on your specific research question(s). Some consider correlations as part of reporting descriptive/summary statistics on all of your variables, which should be routinely presented.

2. Ordinarily, one day difference would not yield a huge impact on just about any variable unless something profound occurred between the two measurements, so I'm a bit unclear as to what Feb 20 vs. Feb 21 actually could represent. Was that the imposition of a work-at-home order, or something similar?

3. It sounds as if a path model (which is really just a series of one or more regressions) might be suitable to address what you're asking, especially if you propose that the two moderator variables are metric (pseudo-continuous and interval strength), and your DV is metric. Likewise, the control variables can be included as antecedents which may influence scores on both occasions (but, in all likelihood, scores on occasion#1 will be related to scores on occasion #2).

Finally, your initial classification of variables as to scale type is pretty reasonable as is. There are some folks who would contest the claim that an averaged score from as few as three Likert-type response items will behave as a metric variable (but you can find plenty of examples of this in the literature; other options would be IRT or factor scores).

Good luck with your work.

I think Dr David Morse has provided a useful answer.

Hello David Morse,

Thank you!

• Can I still use them in linear regression, or should I go for another type?
• Or only linear regression if I create DUMMY variables?
• If only when creating DUMMY variables, I want to create only one for each variable (e.h., responses 2,3,4=1 and ELSE=0). I've read something about the loss of information if I do this, should I be considered about this?

A lot of questions I know, I'm already thankful to get such a quick response on my initial post. I hope it's possible to receive any additional information. Thank you in advance.

Best regards,

Rishi

Correlation , Regression -https://www.bmj.com/about-bmj/resources-readers/publications/statistics-square-one/11-correlation-and-regression.

Hypothesis testing -https://online.stat.psu.edu/statprogram/reviews/statistical-concepts/hypothesis-testing

Hello Rish, You can use my interactive correlation research design tool: http://doctorkarina.net/RES5_5/story_html5.html

I hope it helps. Dr.K

Part I - All Theory

Part II - Correlation in SPSS

Hello again Rish,

You can indeed incorporate categorical or ordinal variables in linear regression models. If a variable has just two levels (e.g., is dichotomous), then code the values as 0/1, or 1/2 and include it. If a variable has three or more levels, then you'll need to recast it as c - 1 dummy variates (where c is the number of levels of the categorical variable); each of these may be coded as 0/1.

For example, if marital status is recorded as single, married, divorced, widowed, then the three dummy variates for any single person in your data set would be 1, 0, 0. For any married person: 0, 1, 0. For a divorced person: 0, 0, 1. And, for a widowed case: 0, 0, 0. In other words, there will be a unique pattern for each level of the categorical variable. All three dummy variates would need to be included in your model to capture the explanatory power associated with that categorical variable.

Good luck with your work.

Correlation shown a relationship between variables and severity of this. Regression shown relationship between variables and predective chance of happen.