Frequently, there are alternative effect sizes we might use to represent the strength of effect in our design. Arguments about which is best may often be subjective. The only essential requirements are that the metric is sensitive the the amount of effect, but that it is not sensitive to sample size. In the case of a Pearson r, one might simply use that r as it is the slope of the scatter-plot between standardized variables. As well, one might use r2 as this is the percentage of variance shared by the two variables correlated.

The R tells you how far the relationship between them, in your case, there is weaknesses. The range is from 0 to 1.

R2 is a more helpful metric of model accuracy, and one that is used more frequently in the literature. It also manifests meaningfulness to the non-statistician. It is simply the percentage of variance in the criterion variable predictable from the model composed of the predictor variables in the sample at hand.

You may also be interested in the cross-validation accuracy of the model. This is a generalization notion -- the degree to which this model is accurate upon application to the population or to another randomly selected sample from that population. The estimator often used for this is incorrect. It is often called the "Adjusted R2. By itself, this is a poor name. The "Adjusted R2" is actually an estimate of the R2 in the population (we would usually use a Greek character here, which I seem to be without herein, so rho2), not of cross-validation performance. There are various formulas to estimate cross-validation performance of a regression model (e.g., Stein-Darlington, Rozeboom, Nicholson-Lord). They are, unfortunately, not common in statistical packages, but I can provide a spreadsheet that calculates them if of interest. In addition, one might choose to pursue a distribution free resampling plan to estimate cross-validation accuracy. David Allen's PRESS estimate fills that bill.