Such great videos on spatial econometrics! I am also learning Geographically Weighted Regressions (GWR), which account for the impact of spatial heterogeneity, or the variations of coefficients across different regions. Would it be possible that you also lecture us on the GWR models?
When both the lag and error models are significant, u can just opt to use the durbin model if you're using R but considering the limitations of options when using the GeoDa software, u can also just use both and compare their AIC values if you just care about what better for the data, but depending on your study, you might just either look at the lag or error.
Yes, as I mentioned, opinions are all over the map on this. But since the lag and error tests tell you little to nothing about the appropriateness of the SLX, Durbin, or Durbin Error models, I favor just running the Durbin model to begin with, as Lesage and Pace suggest. But if your theory suggests a specification, stick with that unless the data are clearly screaming at you to go in another direction. As usual, that is just MY opinion. :)
Hi, great video! Question: I have the same issue as this example (9:15), where my LM test outputs significant robust lags and errors, but insignificant non-robust versions. What is the correct step here? Disregard the robust outputs, follow Luc Anselin and stick to OLS? Or something else (e.g. test SDEM in my case as I expect local effects, following LeSage and Pace)? The video, and later videos on spatial regression, do not explicitly state what to do in this case. Any help would be greatly appreciated!
@@BurkeyAcademy Sir, Could you guide me on how can i do the SARAR (Spatial AutoRegressive-AutoRegressive model) Analysis. Nowadays I am trying to analysis SAR SEM SARAR model for my PhD research. Thank you sir.
Great video! I've got a rookie question about the sample size. For example, if I see a county as a sample, does the number or area of counties matter for spatial regressions.
Not that much- people usually don't bother with it. But, it can matter what the real source of the data is (this is true for spatial or non-spatial regressions). In any kind of regression there is an implicit assumption that the X's are measured without error, and an explicit assumption that the variance of the errors are homoskedastic. So, if data are based on varying sample sizes in counties, say data from the American Community Survey- where samples in some counties will be around 100, and in others many thousands- these measures will have very different standard errors. For X's we call this an "errors in variables" problem. See econ.lse.ac.uk/staff/spischke/ec524/Merr_new.pdf for an overview of the problem and consequences.
Do you have any videos on using GeoDa's space time clustering functions? I have not really understood how the time editor functions, since the time variable doesn't seem to be specified
Hi Pls, explain what to do in my situation. I followed your algorithm. I found both LM-error and LM lag signifacant. But either Robust LM-error or Robust LM-lag were not significant (p values 0.88 and 0.71, respectively). Explain me pls. REGRESSION ---------- SUMMARY OF OUTPUT: ORDINARY LEAST SQUARES ESTIMATION Data set : TMI3 Dependent Variable : TMI Number of Observations: 223 Mean dependent var : 28.2302 Number of Variables : 2 S.D. dependent var : 5.29616 Degrees of Freedom : 221 R-squared : 0.052688 F-statistic : 12.2917 Adjusted R-squared : 0.048402 Prob(F-statistic) : 0.000550846 Sum squared residual: 5925.42 Log likelihood : -682.125 Sigma-square : 26.8119 Akaike info criterion : 1368.25 S.E. of regression : 5.17802 Schwarz criterion : 1375.06 Sigma-square ML : 26.5714 S.E of regression ML: 5.15475 ----------------------------------------------------------------------------- Variable Coefficient Std.Error t-Statistic Probability ----------------------------------------------------------------------------- CONSTANT 34.2168 1.74242 19.6376 0.00000 MGCH -0.0841797 0.0240105 -3.50595 0.00055 ----------------------------------------------------------------------------- REGRESSION DIAGNOSTICS MULTICOLLINEARITY CONDITION NUMBER 9.949618 TEST ON NORMALITY OF ERRORS TEST DF VALUE PROB Jarque-Bera 2 14.8427 0.00060 DIAGNOSTICS FOR HETEROSKEDASTICITY RANDOM COEFFICIENTS TEST DF VALUE PROB Breusch-Pagan test 1 0.1801 0.67131 Koenker-Bassett test 1 0.2537 0.61447 DIAGNOSTICS FOR SPATIAL DEPENDENCE FOR WEIGHT MATRIX : TMI3 (row-standardized weights) TEST MI/DF VALUE PROB Moran's I (error) 0.1074 2.7102 0.00673 Lagrange Multiplier (lag) 1 6.2075 0.01272 Robust LM (lag) 1 0.0199 0.88783 Lagrange Multiplier (error) 1 6.3164 0.01196 Robust LM (error) 1 0.1288 0.71967 Lagrange Multiplier (SARMA) 2 6.3363 0.04208
You should watch my later videos where I show you a much better way to do specification tests- start with the Durbin model instead. The Anselin Method is too restrictive, and is not likely to lead to the best model.
Awesome! I have actually used GeoDa to run spatial error model. I like to use my model for prediction on new dataset. However, I am not sure how to write an equation using the computed coefficients. In fact, not sure if this would be even possible, given that a term related to y variable is also in the predictor variables.
I guess im asking the wrong place but does any of you know of a way to get back into an instagram account? I was stupid lost my password. I love any tips you can give me!
@Kalel Ezequiel Thanks for your reply. I found the site on google and im waiting for the hacking stuff atm. Looks like it's gonna take quite some time so I will get back to you later with my results.
Thanks for the information! I am using Windows, and on the version I had installed I told it to "Check for Updates". When I did, it told me that I had the latest version available. I guess I should not have trusted it! ☺In any case, the version I told you (1.8.16.4) was the one I used, since I have not updated it or reinstalled since that video.