Great question. Here are some thoughts about why simple analytic work shows up as often as it does.

A key audience of the products of agencies like the IBGE are politicians and policy makers. For many of them, presentations of simple descriptive statistics can be challenging enough to interpret for evidence of social progress, new problems, changes from past trends, disparities, and significant findings. In agency anslyses of health statistics, with which I am most familiar, descriptive epididemiology, using simple stratified results, often include projects about discovery of new things: new diseases;, new patterns of outbreaks and spread of diseases; identification of vulnerable populations, and changing patterns of health outcomes.

Such work depends upon credible data sources, effective data collection and reporting systems, and social infrastructure to merely begin the possibility of knowing simple, but important things about a society's emerging health challenges. Such findings enable policy decisions for adjusting social programs and prioritizing allocation of limited social resources to begin to deal with them.

More complex multivariate analyses typically follow these critical discovery and descriptive analyses, which identify variables that first, in simple analyses, demonstrate impact on the findings of interest. Then, having verified and validated those initial findings of what impacts what, more complex forms of modeling and adjusted analyses, like population adjustments, and case mix adjustments, incorporate the variables identified as being important from the earlier discovery work.

Often these more complex analyses are jump started by academics who request access to the agency raw data, and conduct studies of the more advanced kind of inqury you sound interested in conducting, and publish to bring attention to the need for handling decision making more effectively with more factors incorporated into the modeled issues.

An example of identification of simple, important factors impacting outcomes, and then use of those factors in more advanced analysis may be seen here: https://www.researchgate.net/publication/13140640_Developing_quality_of_acute_care_indicators_through_claims_data_augmentation?ev=prf_pub

Article Developing quality of acute care indicators through claims d...

I surely understand the need to make data simple to facilitate the comunication to different kinds of audience. But, as these audiences do not question the methodology of the indicators presented, they will not question the methodology of the indicators presented if they were calculated using modelling approach...

So, official agencies could publish they results, still using income as an exemple, as "income is positively related with school years, even when adjusted by race and gender" or something like this. But these "adjustments" should be done by modelling, not simple stratifications by just one or two variables. Even a pretty naive statistician can see that looking multivariate data in a bivariate form is misleading!

In Brazil, this kind of analysis is used to reinforce every type of discriminations claims. Although I don't completely disagree that some kind of discrimination is present in brazilian society, I disagree that wrong statistics are the best way to show it...

Your comment about what naive statisticians will or will not see makes me sense that the topic of income inequality as it's presented by Brazilian government is a charged topic for you. I do not wish to stoke or challenge those feelings. I did attempt to make a case for beginning and credentialing simple things, and building on that when pursuing more advanced modeling. Perhaps my response offended you. I had no intention fo do so.

Poverty and crime are highly sensitive issues everywhere, and especially as Brazil faces hosting of the World cup very soon. Pacification of the favelas in Rio can accomplish only so much, and perhaps only temporarily. The challenges of poverty and crime appear well beyond the reach of statistical approaches to characterizing them, not the least of which are the questionable data sources for informing them, due in part to the risks in gathering them (doing a census in the favela), much less whether or not simple or complex analyses will better tell their stories.

Your question is important, and if multivariate modeling can improve conditions anywhere, I support it. For me, the translation of analytic findings into improved conditions has taken not only defensible analytic work, and much refinement. It has invariably involved social change efforts, like, publishing findings, speaking to system leaders, attaching economics to findings, and bringing issues to invested communities and partners, and telling stories about indiividuals impacted by the findings that people can relate to.

I found a couple of examples of true stories about poverty and crime in Rio that may communicate much more effectively about what poverty means in Rio's favelas than any government analysis might: a movie, "City of God", and an article that used the viewpoint of a favela resident, found at, http://www.smithsonianmag.com/people-places/A-Look-Into-Brazils-Makeover-of-Rios-Slums-183842911.html?c=y&page=1

I believe this caters for a wide range of audiences, ranging from less technical to more technical. For more detailed/multivariate tables or data one can always use the microdata, in the case of a developed country.

Dear Edwin, I have no issues about income, race, gender, etc. I only took this as example because is a commom ocurrence. My question is more generic and not related to Brazilian reality, but with statistical methods. I just do not understand how delivering a wrong analysed data will turn it into a more understandable information.

Again, people who doesn't know about statistics will not ask how data was worked and people who understand will question data because wrong treatment...

Dear Sandro, I agree to you that official statistics are not fully analyzed. However, that may be a challenge for us researchers, too. Official statistics are in a sense sponsored by citizens. People from different sectors can work together to improve their values.

Two cents from my side:

Given the data collection was done carefully and the infered population is well enough represented by the data, then those effects that are clearly seen without adjustment for covariates are likely to be the strongest ones, also after proper adjustment. When there are limited resources to act on these results, one would anyway tackle the largest effects first.

However, there are some problems with using unadjusted results:

1) relevant effects may not be recognized without adjustment

2) adjuted results may allow for a more focused allocation of resources

3) spurious correlation, collinearities in the predictors may cause "false-positive results" (such problems would become apparent in a multivariate analysis, for example through a variance inflation)

I also think that we won't loose anything by analyzing the data with appropriate multivatiate methods and communicate the adjusted results. I don't think that politicians or managers will be unable to interpret such more differentiated results. For instance, anyone will naturally understand that a low-calory diet may be beneficial for obese persons but not for persons with normal of low body fat - what is already an adjustment of the factor "diet" on the factor "obesity", when a relevant interaction is present (there may be better examples than this one...).

Jochen, thank you! I have raised your last point some comments ago. People get the final information! They don't care about how the results were obtained. If you tell them the raw data or run a fancy statiscal model before, it will make no difference to them...

The problem is even bigger when official statistics comes with a pretend adjustment, like when you say that income is different even to people on the same jobs, because there are many factors that can affect that "same jobs". A manager from microsoft will not have the same income of a manager from a small company...

I asked this question many times in many different forums and never get a objective answer from people who work on official statistics. They don't care to answer? They don't want to answer? They don't know the answer? They don't exist?