Sign of pearson correlation with negative regression coefficient I want to obtain the pearson correlation of "experienced disfluency" and "narrative engagement" from the attached unstandardized coefficients for my meta-analysis.  The paper is from Walter, N., Bilandzic, H., Schwarz, N., & Brooks, J. J. (2021). Metacognitive approach to narrative persuasion: The desirable and undesirable consequences of narrative disfluency. Media Psychology, 24(5), 713-739.
Given that the R square is listed as .23 and the regression coefficient is in negative sign (-.34).  The pearson correlation would then be -.48 (.48 is the square root of R square .23).  Correct?
Also, to double check, the calculation of disfluency in this paper actually use the reverse code of fluency scale.  And since my meta-analysis is about the relationship between narrative engagement and fluency (not disfluency).  The ultimate pearson correlation I should go with should be .48.   Am I on the right track?
Many thanks for the help.

 A: The way you have proposed dealing with signs and their reversal sounds fine to me.
It is hard to know how best to obtain any particular correlation coefficient (r) from the diagram’s r-squareds, which to me are not clearly labeled as to what variables are referred to in each case.  Also, usually one uses r-squared to describe the ability of an entire model – not an individual predictor -- to account for variance in an outcome.
In addition, presumably the coefficients listed in the diagram apply after controlling for (adjusting for, or partialing out) other predictors in the model.  Partial r and zero-order (bivariate) r are not the same!
Then again, I cannot tell exactly what procedure was used here.  In what I have seen online, the full article costs $50.  The abstract (e.g., https://www.tandfonline.com/doi/abs/10.1080/15213269.2020.1789477?journalCode=hmep20) defies convention for quantitative articles in that it does not reveal what statistical methods were used to address any research questions.
Beyond that, the abstract sounds uncomfortably like output from the Postmodern Generator (https://www.elsewhere.org/pomo/).  I don’t feel highly optimistic that, before drawing their quantitative conclusions, Walter, Bilandzic, Schwarz and Brooks paid close attention to the validity and reliability of their variables or to the assumptions behind the multivariate procedure they seem to have used.
