I understand:
- Probs of '(Seriese2-Serise3)/Serise3', 'Serise4', and C are significant enough since they are less than 5%.
- R-squared and Adjusted r square are good enough for my model in predicting 'Serise1': about 89% of 'Serise1' can be predicted.
But I'm not sure:
- Why 'Sum squared resid' is differ from 'Sum squared resid', and how should I interpret this?
- What exactly 'Log-likelihood' means, and if it is ok for 'Log-likelihood' to be this small (-130).
- Is it ok for F-statistics to be so high, even though Prob(F-statistic) is 0.000005?
In general, I'm not certain of my model... Maybe because I'm not that familiar with econometrics but international relations. I tried to express one of the opinions from international relation into econometrics with a simple regression model. I would be really grateful if I can get some help.
Hyeonjun Cho -
Looking at the t-statistics as something like inverter relative standard errors, it appears that your results are strongly dependent on the intercept (constant). That is potentially a problem. The large R-square value is nice, but nothing you can count on. The sample size is small, which could also be a problem. One of the two predictors is a composite of sorts, so I think you need to examine whether or not that is a good idea.
I suggest that you look at a "graphical residual analysis" to check for "fit." You can put the predicted y values on the x-axis of a scatterplot, and the estimated residuals on y-axis. To check for overfitting to a particular data set, a "cross-validation" would be a good idea, but you have limited data, so that is problematic.
Best wishes - Jim
As a learning case, there is no problem. As a research case, it is not usable. Your sample is only 13 observation. No "generalizability" can be made. As for the result print out for various test---as long as you have data input, the software will produce output. The researcher must use good judgment whether adequate data is used. what should be your proper sample size?
I am sure that you will get the detailed answer to your question if you have a look on the relevant chapter of my book and particularly see the pages 53-91 (Dynamic Systems in Management Science. Design, Estimation and Control, Palgrave Macmillan, UK, 2015)
Good luck
Alexis Lazaridis
Without knowing which variables you use and what you want to "explain" it is difficult - if not impossible - to interpret the results. Therefore, only some formal points:
(Series2-Series3)/Series3 is obviously a percentage deviation of S2 from S3. I expect that both have a similar dimension, which means that these are very small amounts compared to the dependent variable with a mean of almost 50000.
S4 seems to have a similar dimension as the dependent variable and a trend in the contrary direction. But it could be that its coefficient is, in essence, due to one or two special data points (because there are so few observations).
The estimated constant is more than double the mean of the dependent variable. I think it is difficult to explain that.
I think these points are sufficient not to accept this regression. You should rethink your hypotheses and what transformations of the data are reasonable. And before going into regressions, you should have a thorough look on the data (e.g. looking at diagrams).
Thank you! James R Knaub Paul Louangrath Alexis Lazaridis Anton Rainer It was really helpful for me to adjust variables! Thank you again!
I cannot read or interpret your results sheet without having details about your variables.In general, the R-squared (.897) is high.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude