The main idea of this thesis is studying the importance of the carry-over effects and interaction effects in statistical models. To investigate it, a hand-milking experiment in Burkina Faso was studied. In many no electricity access countries, such as Burkina Faso, the amount of milk and milk compositions are still highly  relying on hand-milking techniques and milkers. Moreover, the time effects also plays a important role in stockbreeding system. Therefore, falling all effects, carry-over effects and interaction effects into a linear mixed effects model, it is concluded that the carry-over effects of milker and hand-milking techniques cannot be neglected, and the interaction effects among hand-milking techniques, different milkers, days and periods can be substantial.

在本文中，分別就四種不同參數組合（包括六個模式）的二維交叉實驗設計，採用一般線性模式法及二樣本t檢定法予以分析，並探討模式間與分析方法間的異同。此外，在二維重覆測量交叉實驗設計方面，我們也分別以單變量分裂區集變異數分析法及多變量變異數分析法進行探討。 / Four possible parametrizations ( including six models ) are considered in this study to clearify some ambigous issues related to a 2*2 cross-over design. Each model is analyzed using both the GLM procedure and two-sample t test. In addition, we also discuss issues related to the 2*2 repeated measurements cross-over design by using the univariate split-plot and multivariate analysis of variance techniques.

Bioequivalence determines if two drugs are alike. The three kinds of bioequivalence are Average, Population, and Individual Bioequivalence. These Bioequivalence criteria can be evaluated using aggregate and disaggregate methods. Considerable work assessing bioequivalence in a frequentist method exists, but the advantages of Bayesian methods for Bioequivalence have been recently explored. Variance parameters are essential to any of theses existing Bayesian Bioequivalence metrics. Usually, the prior distributions for model parameters use either informative priors or vague priors. The Bioequivalence inference may be sensitive to the prior distribution on the variances. Recently, there have been questions about the routine use of inverse gamma priors for variance parameters. In this paper we examine the effect that changing the prior distribution of the variance parameters has on Bayesian models for assessing Bioequivalence and the carry-over effect. We explore our method with some real data sets from the FDA.

Two studies were conducted to determine the effect of panelist participation frequency and specific aspects of questionnaire design on overall acceptance scoring in consumer central location tests. Regarding participation frequency, research subjects who participate frequently in some survey types are known to provide responses that differ from subjects who participate less frequently, known as panel conditioning. With respect to questionnaire design, overall acceptance (OA) question placement and usage of pre-evaluation instructions (PEI) in questionnaires for food sensory analysis may bias consumers' scores via carry-over effects. To investigate these concerns, data from consumer sensory panels previously conducted at a central location, spanning 11 years and covering a broad range of food product categories, was extracted, compiled, and analyzed. For the first study, data was analyzed to determine evidence of panel conditioning by measuring the effect of participation frequency on mean consumer OA scoring among frequent, moderate, and infrequent participants. Practical significance and occurrence of panel conditioning, defined as mean scoring differences of ≥ 0.50 on a discrete 9-point scale hedonic point, were examined. Results indicate that for overall acceptance, in general, mean scoring differences were not practically significant and did not signify occurrence of panel conditioning. For the second study, OA question placement was studied with categories designated as first (the first evaluation question following demographic questions), after non-gustation questions (immediately following questions that do not require panelists to taste the product), and later (following all other hedonic and just-about-right questions, but occasionally before ranking, open-ended comments, and/or intent to purchase questions). Additionally, each panel was categorized as having or not having PEI in the questionnaire; PEI are instructions that appear immediately before the first evaluation question and show panelists all attributes they will evaluate prior to receiving test samples. Post-panel surveys were administered regarding the self-reported effect of PEI on panelists' evaluation experience. OA scores were analyzed and compared (1) between OA question placement categories and (2) between panels with and without PEI. For most product categories, OA scores tended to be lower when asked later in the questionnaire, suggesting evidence of a carry-over effect. Usage of PEI increased OA scores by 0.10 of a 9-point hedonic scale point, which is not practically significant. Post-panel survey data showed that presence of PEI typically improved the panelists' experience. Using PEI does not appear to introduce a meaningful carry-over effect.