Abstract. The recent growth in organic farming has given rise to the so-called ''conventionalization hypothesis,'' the idea that organic farming is becoming a slightly modified model of conventional agriculture. Using survey data collected from 973 organic farmers in three German regions during the spring of 2004, some implications of the conventionalization hypothesis are tested. Early and late adopters of organic farming are compared concerning farm structure, environmental concern, attitudes to organic farming, and membership in organic-movement organizations. The results indicate that organic farming in the study regions indeed exhibits signs of incipient conventionalization. On average, newer farms are more specialized and slightly larger than established ones and there is a growing proportion of farmers who do not share pro-environmental attitudes. Additionally, a number, albeit small, of very large, highly specialized farms have adopted organic agriculture in the last years. However, the vast majority of organic farmers, new and old ones included, still show a strong pro-environmental orientation.

In order to measure active drag during front crawl swimming a system has been designed, built and tested. A tube (23 m long) with grips is fixed under the water surface and the swimmer crawls on this. At one end of the tube, a force transducer is attached to the wall of the swimming pool. It measures the momentary effective propulsive forces of the hands. During the measurements the subjects' legs are fixed together and supported by a buoy. After filtering and digitizing the electrical force signal, the mean propulsive force over one lane at constant speeds (ranging from about 1 to 2 m s-1) was calculated. The regression equation of the force on the speed turned out to be almost quadratic. At a mean speed of 1.55 m s-1 the mean force was 66.3 N. The accuracy of this force measured on one subject at different days was 4.1 N. The observed force, which is equal to the mean drag force, fits remarkably well with passive drag force values as well as with values calculated for propulsive forces during actual swimming reported in the literature. The use of the system does not interfere to any large extent with normal front crawl swimming; this conclusion is based on results of observations of film by skilled swim coaches. It was concluded that the system provides a good method of studying active drag and its relation to anthropometric variables and swimming technique.

No abstract

B e r i c h t e u n d d i s k u s s i o n e nZusammenfassung: Logit-und Probitregression werden als multivariate Analyseverfahren zur Analyse von dichotomen abhängigen Variablen in den sozialwissenschaften routinemäßig eingesetzt. Beide Verfahren können so interpretiert werden, dass sich aus einer linearen Modellierung einer unbeobachteten Variable y* eine nichtlineare Modellierung der Wahrscheinlichkeiten für y = 1 ergibt. Wir zeigen erstens, dass diese nichtlinearität im Vergleich zu linearen regressionsverfahren zu Problemen bei der interpretation der Modellergebnisse führt. insbesondere die in der logistischen regression häufig verwendeten odds ratios (exponierte Logit-koeffizienten) sind unseres erachtens problematisch. stattdessen empfehlen wir neben graphischen interpretationshilfen die Verwendung von (korrigierten) durchschnittlich marginalen effekten (AMe). Zweitens zeigen wir anhand einer serie von Monte-carlo-simulationen, dass die üblichen regressionskoeffizienten bei Logit-und Probitanalysen nicht zwischen verschachtelten Modellen verglichen werden können. da in den sozialwissenschaften bei der Modellbildung jedoch häufig schrittweise vorgegangen wird, wäre ein Verfahren, das einen validen Vergleich von effektstärken zwischen den Modellen erlaubt, sehr nützlich. Wie wir anhand unserer simulationsstudie zeigen, führen durchschnittlich marginale effekte und koeffizienten, die nach dem Vorschlag von karlson et al. ( Sociological Methodology 42, 2012) korrigiert wurden, in sehr verschiedenen situationen zu gültigen ergebnissen. y*-standardisierte koeffizienten sind für einen Modellvergleich hingegen weniger geeignet und koeffizienten eines linearen Wahrscheinlichkeitsmodells sollten ausschließlich bei normalverteilten Variablen verwendet werden.

No abstract