Census workers are settling for incomplete long questionnaires in the final push of the summer collection period, raising concerns the data will be even more compromised than originally feared.Statscan settles for incomplete long-form surveys in 2011 census - The Globe and Mail (“Canadian Conservatives Contra Canadian Census”, Round 2, still winning)
The ESS website published this bit of news in the summer:
ESS4 and ESS5 Corrected design weights for Israel
All Israeli respondents in ESS4 and ESS5 have been assigned wrong design weights. New files containing the corrected variable DWEIGHT for Israel have been made available from the ESS4 and ESS5 Data download sections.
This is directly relevant to this course do-file on torture in Israel. The
ess0810 teaching dataset used in the do-file is based on the ESS Round 4 Edition 4.1 and the ESS Round 5 Edition 3.0. These editions do not include the corrected Israeli weights. Neither does the current cumulative ESS dataset that one can download in SPSS format by looking at the source code of the ESS website.
Here’s a script that diagnoses the issue and creates the dataset used to patch the teaching material with the corrected weights. Note that the difference between the original and corrected design weights is not at all trivial:
It’s the second time that the course material is affected by an ESS update. The first update affected education measurements. These corrections are likely to also affect results shown in the documentation and training modules, such as this table of Internet use.
Related: the first release of ESS6 is planned for late October this year.
It’s common to misinterpret that low correlation as a sign of poor prediction. Again – thousands of research studies have confirmed the predictive validity of tests of cognitive ability for job performance. Google may well say, “Not here.” – but they cannot (and did not) say, “Not anywhere.”Test scores and grades predict job performance (but maybe not at Google) « Statistical Modeling, Causal Inference, and Social Science Statistical Modeling, Causal Inference, and Social Science
In the control group, the authors find what Bartels, Nyhan and Reifler found: There are big partisan gaps in the accuracy of responses. …. For example, Republicans were likelier than Democrats to correctly state that U.S. casualties in Iraq fell from 2007 to 2008, and Democrats were likelier than Republicans to correctly state that unemployment and inflation rose under Bush’s presidency.Survey respondents are lying, not ignorant | Stats Chat
Worthwhile Canadian Initiative: How should economists respond to the National Household Survey release?
The voluntary nature of the NHS will compromise the quality of the data collected. For example, the National Household Survey asks people about their religious beliefs. Yet religion has a strong influence on volunteering and civic engagement. The National Survey of Giving Volunteering and Participating found that people who are religious volunteer more. Other studies argue that some religions discourage participation in civic life - but reinforce the message that religion matters.
There is, therefore, likely to be a correlation between a person’s willingness to fill out the National Household Survey and the strength of their religious beliefs. How, then, can we possibly know whether the NHS information on the religious beliefs of the Canadian population is accurate?
The previous post and the one before that mentioned that plotting survey data, which often contains ordinal or low-dimensional nominal data, can take many Stata options. I have started working on a wrapper for Nick Cox’s
catplot command to bring down the code to one-line commands that produce graphs like the following examples:
svyplot marital, ymax(60)
The example above is close to the default
catplot output with one variable. With two variables, I have tried to implement degrading colors as shown in the work of the Oxford Internet Institute:
svyplot health race, asc red ymax(60)
The wrapper uses
blues (default) for the color gradients, which can be
ascending (default) or
ymax option controls the height of the graph, which is
100 by default, in order to fit stacked bars:
svyplot happy polviews, des stack angle(25) scheme(burd3)
svyplot inequal3 race, asc hor stack scheme(burd5)
This final example shows stacked horizontal bars. The wrapper code probably won’t behave well with
recast(dot) and three-variable arrangements, even though both are supported.
What is this I don’t even—
We use data from a unique survey of members of drug-trafficking gangs in favelas (slums) of Rio de Janeiro, Brazil, to characterize drug-trafficking jobs and study the selection into gangs, analyzing what distinguishes gang-members from other youth living in favelas.Your probability of death if you join a Brazilian gang is 20% over two years | Chris Blattman