This is what some researchers suggest, and also some of my previous studies came up with results consistent with this notion. I will be glad to have some specific reference to this phenomenon. Any clue? Thanks
Yes could be, but is not of danger, as according to Halcoussis (2005, 58) "understanding econometrics", There is no standards for R square, on that says, for example, "An R square larger than 0.75 (or any other number) means the model is good."
Typically, the R square is higher in time -series regressions than in cross-sectional regressions. The area of study is important also, if changes in the dependent variable are hard to explain, then 0.40might be a great R square, but if the dependent variable is easily predicted,an R square of 0.80 may indicate a poor fit.
What follows is copied and pasted with some modifications from an earlier answer of mine...........
When analyzing individual (not aggregated) data low values below 5% are not unusual - you have to decide is it practically useful and have the assumptions behind the analysis been met. Individuals are typically very heterogonous in their attitudes, actions and behaviours.
I am reminded of a famous clinical trial of the effect of taking aspirin on heart attack - the odds ratio was so dramatic that the trial was stopped and placebo group advised to take aspirin. And yet the odds ratio of a heart attack for placebo compared to taking aspirin was, a rather a lowly 1.83 and the R2 was a puny 0.0011; yet this was sufficient for action.
Your arguments are strengthened if you are testing a relationship and have not gone on a fishing expedition and you have tried to take account of theoretically relevant confounders. Epidemiology has gone to some extent from 'what are the causes of this outcome?' to 'does this potential cause have an effect?.
I would also add if you are modeling binary(0 and 1) outcomes it is exceedingly difficult to achieve high R2 values as the predicted probability values are not very likely to be exactly 1 and 0!
Finally we have to accept that there are outcomes where chance does genuinely part a large part so we now have evidence that luck plays a bigger part in some cancers than genes and lifestyle; see
So for me it is theory, focused question and the size of the slope term, care about confounding and careful evaluation of the model rather than just R2.
see other postings at
R-squared values between 1 and 5% in linear regression social science? - ResearchGate. Available from: https://www.researchgate.net/post/R-squared_values_between_1_and_5_in_linear_regression_social_science [accessed Sep 1, 2015].
R Square means the % of explained variance in the dependent variable. When trying to test for the impact of certain limited number of variables on a human related outcome - be it attitude or action, one cannot expect to have the majority of variance explained - it is not physics or math. A model can only test few variables out of the many possible antecedents (and it cannot test luck and chance events). So in the behavioural sciences, .5 is quite good, and even .3 or .4 is a fair level of explained variance.
I agree that an R-sq of 50% is relatively high for the social sciences. Many of the variables we test are subject to measurement error (which will attenuate correlations) and many of our theories are overly generally (make predictions as if they apply to everyone in a population). So I personally think that an an R-sq of around 30% is pretty good in most cases.
I truly appreciate all the answers. Seems like there is a general agreement among researchers on this issue. However, could you please cite any reference in this regard? Thanks.
Whatever good statistical book has this information about R...ex Morris Hamburg, Agresti, Bayo laval, Glass Hopkins; Sharma, ....and also books about applied statistics to psycology.
If this is not helpuf, please tell me and i would be more precise (pages, ets..books...)
In social science research, sometimes it is quite difficult to have higher R-square value. However, even when you get low R-square value, it does not reflect that your study will not be accepted. Practically and realistically, usually we experience relatively lower R-square value between .3 to .5, which is quite OK in social science research.
To emphasize what Yehuda has said, r-square is the percent of variance in the dependent variable explained by the independent variable. So, an r-square of .5 (unless you meant .05) says that 50% of this variance is explained (not 5%), and that's outstanding in the social sciences!
Lots of good points, but it is worth adding a few further caveats. First, R^2 is the variance accounted for in the sample by your model. There is a tendency to interpret as a population estimate (which it is not). Second, it is a point estimate and in small or noisy samples may be a very bad one. Third, in addition to sampling error, all sorts of things can influence R^2 that have nothing to do with the theoretical or practical questions you are interested in. For example, range restriction, sample characteristics, ceiling effects, floor effects, measurement error, and small sample bias all influence R^2. Finally, R^2 does not generalise well to certain kinds of models (e.g., multilevel models, generalised linear models etc.) and thus it can be very misleading - especially when variance is a function of the variables in the model.
According to Hair et al. (2010), although the acceptable level of R square value depends on the research context. Meyer, et al. ((2013) proposed an R square value of 0.1as a minimum acceptable level. Chin (1998), suggested that R-square values of .19, .33, and .67in PLS-SEM can be considered as weak, moderate, and substantial respectively.
I haven't read Hair et al. (2010) but any fixed point such as 0.10 is nonsense unless it refers to a very specific context (and even then I'm not a fan of fixed cut-offs). Kelvin's answer gives some examples. Kelvin mentioned the aspirin example, but there are others (such as the Salk vaccine trial) where tiny fractions of a percent of R^2 represent effects of huge practical importance.
I'm also unsure that PLS-SEM is ever a good idea (or indeed that it is a form of SEM):
The first question is to know if you are speaking of 0,5 percent or 0,5 as 50%. If it is the first it is really a small explained percentage, if you are speaking of 50% is rather good for the social sciencies. The other answers explain the rest...
For me as an individual participant, below 0.5 (of 1 max.) would be very common. Attitude vs. Facebook (0 - too many ads and too much blabla), G+ (0.5 - no ads but not much traffic but useful insights occasionally).
I am currently analysying data for a human behaviour study. Are there any references which indicate acceptable R2 values for human behaviour studies, given the unpredictability of same
Mohd Hasanur Raihan Joarder sorry for re entering an old post. In your post above you refer to a journal Meyer et al. 2013. Do you have a title for the journal as it might be a useful reference for me.