Agriculture/biometrics
Presenter: Fatima Batool
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
A Generic Randomization Device for Estimation of Mean of Quantitative Sensitive Response Variable
This study concerns about modeling of sensitive phenomenon frequently occurring in behavioral, political, medical and sociological studies among others. Designing surveys for highly sensitive issues for instance homosexuality, commercial sex exploitation, drug additions or induced abortions are hard due to social desirability bias. Usual survey techniques do not apply here due to response and non response bias risk. Therefore, objective in such studies is to gain respondent cooperation and gather truthful response as such question haunts respondents privacy. We have designed a generic randomization device (RD) and propose 12 models for the estimation of mean of quantitative sensitive characteristic. The challenge in proposing this RD was to develop a setup that makes use of additive, multiplicative and both additive and multiplicative scrambled response with blank card strategy. We have also proposed quantification of Perri (2008) model and explored the statistical properties of all proposed models. Since 12 models are proposed and for each there are various possibilities of parametric values for study, scrambling variables and design probabilities thus, numerical comparison is very tricky and tedious. Extensive simulation studies have been done to explore the efficiency behaviors of the proposed models with each other and with existing models. To summarize these studies important findings are presented in tabular and graphical forms. For real survey planning such choices of scrambling variables are suggested which will reduce the burden of computation at estimation stage for researchers. All proposed models are superior in terms of efficiency and it is interesting to note that various existing models can now be viewed as the special cases of proposed RD.
Agriculture/biometrics
Presenter: KIRIKO ABDALLAH
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
GRAIN DEMAND AND CONSUMPTION IN SELECTED DISTRICTS OF UGANDA. THE CASE OF MAIZE, BEANS, GROUNDNUTS AND RICE.
This study examined grain Demand and consumption in Uganda with special focus on Lira, Kabale and Mbale districts to inform food policy formulation. The study, among other things, sought to determine consumption expenditure patterns for selected grains and to provide demand elasticities for maize, beans, ground nuts and rice. The study uses cross sectional data of the Uganda National Household Survey 2012/2013 collected by Uganda Bureau of Statistics. The study analyzed a sample of 200 households selected based on the districts of interest and whether the households consumed the 4 commodities under investigation. The data were merged and tested for homogeneity and symmetry using the Wald test. Results of the study on food grain consumption expenditure patterns show that maize flour and beans are the most consumed food grains. Econometric results of the LA/AIDS models show that household size, occupation of the household head, household expenditures on the food grains and per capita expenditure were statistically significant at the 5% level. Considering household consumption expenditures on the four grains. In general, the results show that 62.5% of the total household food budget is spent on the four food grains under investigation which implies that the four food grains are important in the dietary system of households. Marshallian own price elasticities for dry beans (-0.89), maize flour (-0.25), rice (-1.13) and pounded groundnuts (-0.72) are consistent with economic theory. The expenditure elasticities for dry beans (0.60), maize flour (0.72) and rice (0.98) show they are normal food commodities. Pounded groundnuts are luxury food commodities (?g =1.97). Hicksian cross price elasticities show that beans can be substituted for pounded groundnuts and maize flour can be substituted for rice. The estimates of cross price elasticity indicate that substitution effects of price change were not quite strong. Simultaneously no systematic differences in the absolute magnitudes of the expenditure elasticity and own price elasticity were found. This implies that a combination of income and price policies may be more effective in influencing consumption patterns of households than those based solely on an individuality basis without taking into consideration the other factor.
Agriculture/biometrics
Presenter: Jhonny Demey
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
In silico selection for resistant genes of diseases and endophytic fungi in Theobroma cacao: a K-Tables Analysis approach
The increase of public data of microarrays available has gained increasing importance in plant research. These have enabled the opportunity of study simultaneousy, based on multiple datasets, the expression levels of thousands of genes over the effects of certain treatments, diseases, and developmental stages on gene expression. This has turn out to be a promising approach for analysing and interpreting genome-wide association studies and gene set analysis that are useful to comprehension of biological processes underlying to more important plants diseases. However, current statistical methodologies for gene set analysis based on multiple datasets are still in an early stage of development, they are mostly based on classical statistical methods, since the joint analysis of the subspaces that generate multiple datasets are not simple. The K tables analysis have been developed to handle these problems. We illustrate the proposed approach using multiple microarray gene expression datasets of three studies asociated to diseases and fungi fungal of Theobroma cacao plants.
Agriculture/biometrics
Presenter: Perumal Venkatesan
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
Rough Set Based Feature Selection Method for High Dimensional Biological Data
Pattern recognition problems require selection of a subset of variables or features to represent the patterns to be classified. In many biological applications, there is a need to reduce the number of features without significantly degrading the performance of the system. The performance of any classifier is sensitive to the choice of the features used to construct the good classifier for high dimension data that are inherently noisy. In this work, an efficient feature selection method for finding and selecting informative features in high dimension data which maximum the classification accuracy is proposed using the rough set theory. This has also the potential to identify informative features combinations for classification. The classification accuracy from the machine learning classifiers are used for comparison. Experimental results with benchmark datasets show the usefulness of the proposed approach for high dimensional biological data.
Agriculture/biometrics
Presenter: SEONWOO KIM
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
Comparison of positive predictive value of a number of diagnostic procedures – An extension to three diagnostic results
Diagnostic procedures are mostly used in the detection of a particular disease, and each procedure indicates the presence or absence of disease in an individual. Positive predictive value, one of measures for the effectiveness of a diagnostic procedure, is calculated as the proportion of the diseased persons amongst individuals of disease detected. For this binary result of diagnosis, positive predictive value of a number of diagnostic procedures applied to the same persons can be compared using Chi-square statistic (Bennett 1972; Biometrics). But, for like the treatment of tumor, it is important to know the location of tumor in advance by the diagnostic procedure, and it is possible to be three diagnostic results, no cancer (a), cancer but indicating incorrect location (b), correct diagnosis of tumor location (c). For this situation, it is of interest how correctly the procedure indicates tumor location when it diagnoses cancer. In this case, positive predictive value can be defined to be the ratio of the number of cases (c) among cancer patients to the sum of the number of cases of (b) and of (c). To compare this kind of positive predictive value of diagnostic procedures, test statistic is developed as an extension of test statistic of binary diagnostic result. Simulation results for evaluation of nominal level and power are presented, and real example data is applied to new test statistic.
