The Sociaal-wetenschappelijke Sectie van de Vereniging Voor Statistiek en Operations Research
(SWS-VVS; Netherlands Statistical Society; Social Sciences Division) promotes the use of new and sound statistical methods
in the social and behavioral sciences.
The SWS-VVS appeals to both statisticians working in the social and behavioral sciences
and applied researchers in the social and behavioral sciences.
The SWS-VVS organizes two annual meetings on a statistical topic, and contributes to the Dag voor Statistiek en Besliskunde.
Members of the SWS-VVS receive the ISI journal Statistica Neerlandica, the VVS newsletter (in Dutch), and the non-specialist
journal StatOr (in Dutch). Members have free entrance to all meetings organized by the VVS.
Moreover, members of the SWS-VVS allow the preservation of the only professional association for
methodologists and statisticians working in the social and behavioral sciences.
Najaarsbijeenkomst 2011
|
Nulhypothesetoetsing in de sociale wetenschappen
|
Op maandag 12 december 2011 organiseert de Sociaalwetenschappelijke Sectie van de Vereniging voor Statistiek een lezingenmiddag en discussiebijeenkomst over Nulhypothesetoetsing in de sociale wetenschappen.
Locatie : Zaal 1.01, Gebouw M van de Universiteit van Amsterdam, Plantage Muidergracht 12, Amsterdam
Deelname is gratis
Aanmelden kan uiterlijk 8 december d.m.v. het sturen van een email naar j.m.wicherts@uva.nl met als titel "lezingenmiddag".
Het programma is als volgt:
Lezingen:
Onderzoekers gebruiken significantietoetsing reflexmatig, opportunistisch en zonder veel kennis.
Rink Hoekstra
Rijskuniversiteit Groningen
Al sinds het begin van de 20e eeuw wordt nulhypothese-significantietoetsing (NHST) hevig bekritiseerd. Het zou onder andere antwoord geven op de verkeerde vraag, en mensen aanzetten tot het te dichotoom beantwoorden van hun vraagstellingen. De discussie over NHST heeft tot enkele veranderingen in voorschriften van onder andere de APA Manual geleid, maar op onderzoekers in de praktijk lijken deze weinig invloed gehad te hebben. Zowel in artikelen als op de werkvloer lijkt NHST nog altijd reflexmatig gebruikt te worden: alternatieven als betrouwbaarheidsintervallen worden zelden overwogen. Daarbij worden ernstige interpretatiefouten gemaakt, waarbij dichotoom denken veel voorkomt : nulhypotheses worden vaak geaccepteerd, en een significant effect wordt regelmatig als onweerlegbaar bewijs voor het bestaan van een populatie-effect gezien. Daarnaast worden effectgroottes zelden geïnterpreteerd. Interviews met onderzoekers suggereren dat een ernstig gebrek aan kennis, en een opportunistische houding ten aanzien van onderzoek in het algemeen en statistiek in het bijzonder belangrijke factoren vormen voor dit gedrag.
Human factors in statistics
Jelte Wicherts
Universiteit van Amsterdam
In the debate concerning null-hypothesis significance testing (NHST) little attention has been paid to human factors in statistics. I argue here that conducting statistical analyses of research data and reporting of statistical results is prone to human error and that researcher's statistical decision-making is often directed at rejecting the null hypothesis at alpha = .05. As a result, human factors introduce a special type of bias that is evident from the disproportionate frequency of p-values just below the .05 threshold. I highlight the high prevalence of errors in the reporting of results from NHST and show that both these errors and the height of p-values are associated with researchers' willingness to share data for reanalysis. I contend that NHST remains popular in no small part because it serves the researcher's need for closure.
Worth no more than a bare mention? An objective Bayesian analysis of the evidential impact of the p value
Eric-Jan Wagenmakers
Universiteit van Amsterdam
In many fields of study, the evidence against the null hypothesis is summarized by the p value. By decree, a p value lower then .05 allows researchers to reject the null hypothesis. Here we present Bayesian analyses that quantify the evidential impact of the p value – that is, for a given p value, how much more likely are the observed data under the null hypothesis vis-a-vis the alternative hypothesis? In Bayesian analyses, one usually has to commit to a specific form of the alternative hypothesis, that is, one has to formalize what effect sizes can be expected under the alternative hypothesis. This subjective specification of expected effect sizes can be avoided by using the specification that is most generous toward the alternative hypothesis. This upper bound on the evidence is objective in the sense that it defines the maximum amount of evidence that can possibly be obtained against the null hypothesis (Edwards et al., 1963; Sellke et al., 2001). A comparison between p values and the upper bound on the evidence yields a conclusion that is truly shocking: For all p values greater than .037 the evidence against the null hypothesis is, at the very best, "worth no more than a bare mention". I illustrate by example how the reliance on "p < .05" dramatically overstates the evidence against the null hypothesis.
Bayes as a golden solution for all your modeling issues?
Rens van de Schoot
Universiteit Utrecht
Bayesian statistics are becoming more and more popular in applied papers to answer the research question at hand. As Kruschke (2011a, p. 272) states in a special issue of Perspectives on Psychological Science: "[...] whereas the 20th century was dominated by NHST [null hypothesis significance testing], the 21st century is becoming Bayesian (as forecast by Lindley, 1975)." This is especially due to the availability of Bayesian estimation methods in popular software like WinBUGS, MlWiN, AMOS or Mplus v6.x. Besides, many researchers switch to Bayesian statistics because it can deal with some commonly encountered problems in orthodox statistics or just because they like it. Advantages of Bayesian statistics over orthodox statistics are well documented in statistics as well as in psychology. In this presentation I will briefly highlight some arguments, namely that:
- The null hypothesis (if you would be interested in it) can be accepted and not only rejected as in NHST. In fact any hypothesis can be evaluated and provided with a measure of support
- The probability of a theory being true given data, i.e., P(theory | data), is estimated and that is what many of us would like to know. This is the inverse of P(data | theory), which is what orthodox statistics tells us
- Credibility intervals, the Bayesian equivalent to a confidence interval, are actually the probability that a certain parameter is in between two numbers, which is not the definition of a confidence interval.
- Bayesian estimation always results in admissible parameter estimates.
- Previous research results can be used in the current data analysis.
Finally, I will argue that switching to Bayesian statistics should not be done carelessness and I will demonstrate some aspects that might be considered the "achilles heel" of Bayesian statistics.
Eerdere bijeenkomsten:
| Board of the Sociaal Wetenschappelijke Sectie |
Het huishoudelijk reglement van de SWS
In de sectievergadering tijdens de Statistische Dag 2011 is het nieuwe reglement aangenomen.
Laatste wijziging 28-10-2011