Statistical inference, exchangeability and multilevel approach

Courgeau, Daniel

doi:10.4000/msh.6793

Cited by 28 publications

(4 citation statements)

References 15 publications

(13 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…But subjective probability allows only very approximate individual forecasts, for people have many other characteristics besides the ones analyzed, and those other characteristics can strongly modify the forecast (Courgeau 2007b We could extend the analysis in the previous chapter to the other social sciences. Similarly, when we want to address a paleodemographic problem, with a limited amount of data, the continued recourse to objective probability yields results that are often totally erroneous and even impossible to accept.…”

Section: Resultsmentioning

confidence: 99%

“…Multilevel analysis, as well, was to draw on Bayesian methods (Goldstein 2003 ) . We have chosen our examples from the fi eld of education science (Rouanet et al 2002 ;Courgeau 2007b ) and of testimonies, from Hooper ( 1699 ) to Dempster ( 1967 ) . Instead, we shall describe one application to multilevel analysis-in particular, the subjectivist statistical inference that the methods allow with the aid of the concept of exchangeability.…”

Section: Applications To Social Sciencementioning

confidence: 99%

“…The MCMC method derives its name from the following notion: to produce acceptable approximations of integrals and other functions that depend on a distribution of interest, it is suffi cient to generate a Markov chain whose limit distribution is the distribution of interest (Robert 2006 ) . We shall use the Bootstrap option only (for the results with the MCMC option, see Courgeau 2007b ) . We have assumed here that a binomial model was well suited, as we can show in a non-Bayesian example.…”

Section: Applications To Social Sciencementioning

confidence: 99%

“…Of course the Bayesian approach allows an even more satisfactory multilevel analysis (Goldstein 2003 ;Courgeau 2007b ;Draper 2008 ) , as in the case of eventhistory analysis. In particular, when the number of units in one of the aggregation levels is small, the maximum-likelihood estimation of asymptotic standard deviations may be heavily biased as a result, and the maximum may actually be negative (Draper 2008 ) .…”

Section: Return Of Dispersion In the Event-history And Multilevel Appmentioning

confidence: 99%

See 3 more Smart Citations

Probability and Social Science

Courgeau

2012

Self Cite

View full text Add to dashboard Cite

This Book Series is devoted to examining and solving the major methodological problems social sciences are facing. Take for example the gap between empirical and theoretical research, the explanatory power of models, the relevance of multilevel analysis, the weakness of cumulative knowledge, the role of ordinary knowledge in the research process, or the place which should be reserved to "time, change and history" when explaining social facts. These problems are well known and yet they are seldom treated in depth in scientifi c literature because of their general nature. So that these problems may be examined and solutions found, the series prompts and fosters the setting-up of international multidisciplinary research teams, and it is work by these teams that appears in the Book Series. The series can also host books produced by a single author which follow the same objectives. Proposals for manuscripts and plans for collective books will be carefully examined. The epistemological scope of these methodological problems is obvious and resorting to Philosophy of Science becomes a necessity. The main objective of the Series remains however the methodological solutions that can be applied to the problems in hand. Therefore the books of the Series are closely connected to the research practices.

show abstract

Section: Resultsmentioning

confidence: 99%