Reporting Data with Estimates and SE

What do you think?

When reporting data in an article, a usual way is to show the mean values +/- the standard errors in a table. For Example, when you measure Abundance of pigeons in parks of five different cities, with each city having 30 parks. You would condense the data and show the MEAN +/- SE for each of the five towns.

However, if you analyse your data using GLM with i.e. poisson or negative binomial distribution. And furthermore you do not only analyse Town as explanator but also covariables, like..whatever, Rainfall, Temperature, amounts of trees in the park, etc. Is it still justifiable to condense the data by showing MEAN and SE values for the towns, or would you expect Estimators +/- SE from the GLMs in the reported data table? Because for me it seems that the MEAN is not very representative when the data are not normally distributed. Or do I mix up two different things? Reporting Data on the one hand and reporting Analysis Results on the other?

• It is unclear what you are asking. Do you mean reporting mean and SE of regression parameters? If so than it seems that you are thinking of their posterior distribution and Bayesian estimation... On another hand, I'm not sure if you are not after some kind of hierarchical model. Could you edit your question to clarify what do you mean? – Tim Apr 23 '16 at 9:34
• If you are going to model using a negative binomial model presumably with a log link is reporting on the original scale a good idea? Would it not be better to report coefficients on the log scale or means using geometric means? If you want coefficients on the original scale then a confidence interval would be better than standard error as the interval is not symmetric on the original scale. – mdewey Apr 23 '16 at 12:37
• Yes, that's the question. I read a lot of paper, about field ecology investigations of abundances of animals. Normally insights on the data are given by showing summary tables with MEAN +/- SE of many, many taxa for the investigated fields, or classes of fields. However, if negative binomial or poisson regresssion is used to analyse differences between the fields/classes later on, you would say that these kind of tables are not adequate? – Pharcyde Apr 23 '16 at 13:05