0

this may be a stupid question, but I have run a mixed effects model in R and the estimates, Standard Error and dfs are are all numbers such as -2.115e-01. I've only seen these in p-values before and I'm not sure to report them when they have no clear relationbship to the original scale (Bark scale ranging from -4 to +2). What is the best way to discuss these findings?

Thanks

tryskew8 = lmer(skewness~ respondent + respondent : duration+ (1|word) + (1|video), REML = F , clean)
summary(tryskew8)

Fixed effects:
                                 Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)                    -2.115e-01  1.775e-01  3.844e+01  -1.192 0.240738    
respondentgay.referee          -4.270e-01  2.876e-01  4.355e+01  -1.485 0.144834    
respondentcoming.out           -1.412e+00  3.111e-01  2.465e+01  -4.538 0.000127 ***
respondentout                  -7.549e-01  3.069e-01  2.335e+01  -2.460 0.021703 *  
respondentphil.in:duration      3.125e-04  1.411e-03  3.823e+02   0.222 0.824788    
respondentgay.referee:duration -2.118e-03  1.950e-03  3.873e+02  -1.086 0.277937    
respondentcoming.out:duration   2.144e-03  1.837e-03  3.832e+02   1.167 0.243932    
respondentout:duration          2.449e-03  1.626e-03  3.769e+02   1.506 0.132956
kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Brad
  • 1
  • Your significant effects are not that small compared to a range from -4 to 2, are they? – jay.sf May 24 '20 at 17:48
  • 1
    Maybe it's just that the scientific notation is crossing you up. "-2.115e-01" indicates -0.2115: just move the decimal point left (if "-") or right (if "+") the number of places indicated after the "e". You can also set R to avoid that notation -- https://stackoverflow.com/questions/9397664/force-r-not-to-use-exponential-notation-e-g-e10#9397821. – rolando2 May 24 '20 at 18:10
  • 2
    Your model seems to include an interaction term involving `duration` without a corresponding term for `duration` on its own. That's almost never a good idea, and can affect the values and interpretations of the interaction coefficients. See [this page](https://stats.stackexchange.com/q/11009/28500) for example. Also, think carefully about your choice to use [maximum likelihood instead of REML](https://stats.stackexchange.com/q/48671/28500) for fitting for model to the data unless you have a very large study. – EdM May 24 '20 at 18:12
  • Excellent, thanks very much! Now the scientific notation makes sense, I've also added a term for duration to the model – Brad May 26 '20 at 11:08

0 Answers0