Dear all Stata users,
I am doing a meta-analysis about the relation between tax fraud and social responsibility of companies. I have already found and read in Stata v16 about meta regress command which is used to perform both fixed and random effcets meta analysis. However I would like to start obtaining regression outcomes from using a more "traditional" estimators (weighted least squares using as weights the inverse variance).
These "traditional" meta analytical estimators may also be both fixed or random, depending on the assumptions we make about the residual heterogeneity of the effect sizes. In this regard, in contrast to traditional fixed effects meta-analysis, random effects explicitly allows for heterogeneity of primary effect size estimates beyond sampling error (εi). This means, as Feld and Heckemeyer (2011) mention, that the true unobserved effect size is not assumed fixed but supposed to contain a study-specific random component (μi). So in RE meta analysis, the weights relies on the inverse of both variance of sampling error as well as the study-specific random component - 1/(𝜎𝜀^2+𝜎𝜇^2). The εi is the standard error of the estimated effect size usually given by the original studies, however μi is not.
My question for which I kindly ask for your help/direction is how to estimate the variance of the study-specific random component (𝜎𝜇^2). In paper of Feld and Heckmeyer (2011), the authors say that they estimate it by using residual maximum likelihood estimation technique, however I am struglling how to implement this. Is there anyone who can help in this query?
Thank you very much in advance for your help. Best regards,
Mário
I am doing a meta-analysis about the relation between tax fraud and social responsibility of companies. I have already found and read in Stata v16 about meta regress command which is used to perform both fixed and random effcets meta analysis. However I would like to start obtaining regression outcomes from using a more "traditional" estimators (weighted least squares using as weights the inverse variance).
These "traditional" meta analytical estimators may also be both fixed or random, depending on the assumptions we make about the residual heterogeneity of the effect sizes. In this regard, in contrast to traditional fixed effects meta-analysis, random effects explicitly allows for heterogeneity of primary effect size estimates beyond sampling error (εi). This means, as Feld and Heckemeyer (2011) mention, that the true unobserved effect size is not assumed fixed but supposed to contain a study-specific random component (μi). So in RE meta analysis, the weights relies on the inverse of both variance of sampling error as well as the study-specific random component - 1/(𝜎𝜀^2+𝜎𝜇^2). The εi is the standard error of the estimated effect size usually given by the original studies, however μi is not.
My question for which I kindly ask for your help/direction is how to estimate the variance of the study-specific random component (𝜎𝜇^2). In paper of Feld and Heckmeyer (2011), the authors say that they estimate it by using residual maximum likelihood estimation technique, however I am struglling how to implement this. Is there anyone who can help in this query?
Thank you very much in advance for your help. Best regards,
Mário
