Explore chapters and articles related to this topic
Let’s Find Out
Published in S. Kanimozhi Suguna, M. Dhivya, Sara Paiva, Artificial Intelligence (AI), 2021
Jayden Khakurel, Indu Manimaran, Jari Porras
Quantitative data collected from the two sessions were analyzed using the statistical data analysis language R and the descriptive statistical analysis functions available in R core (R Core Team 2017) and the psych library (Revelle 2017). We first used the Mann–Whitney U test (Wohlin et al. 2012) to analyze the difference in distributions between the data sets. A continuity correction was enabled to compensate for non-continuous variables (Bergmann and Ludbrook 2000). The Bonferroni correction was used to adjust the p-value to compensate for the family-wise error rate in multiple comparisons (Abdi 2007). We calculated the effect size r using the guidelines by Tofan et al. (2016) for the Mann–Whitney U test. We evaluated the effect size as proposed by Cohen (1994): in r, a large effect is 0.5, a medium effect is 0.3, and a small effect is 0.1.
Dual Energy Computed Tomography for Lung Cancer Diagnosis and Characterization
Published in Ayman El-Baz, Jasjit S. Suri, Lung Imaging and CADx, 2019
Victor Gonzalez-Perez, Estanislao Arana, David Moratal
The licensed statistical software package SPSS 20 (IBM, Somers, NY, USA) was used in the statistical analysis. A bivariate analysis was performed using the Mann-Whitney U test to distinguish tumor characteristics (benign or malignant lesion, ADC vs. SCC and necrotic status). Significance was set at p = 0.05, and Bonferroni adjustment was used to correct the significance level due to multiple comparisons. Receiver operating characteristic (ROC) curves were generated, and diagnostic capability was determined by calculating the area under the ROC curve (AUC). According to the AUC value [31], parameters can be considered poor (AUC < 0.6), fair (0.6 < AUC < 0.75), good (0.75–0.9), very good (0.9–0.97), and excellent (0.97–1) to differentiate between tumor characteristics.
Introductory Statistical Experimental Designs
Published in Jiro Nagatomi, Eno Essien Ebong, Mechanobiology Handbook, 2018
Julia L. Sharp, Patrick D. Gerard
Oftentimes, we would like to compare the means of more than two levels of a factor. The analysis for the designed experiment with more than two levels of a factor has several advantages, including ease of the analysis, and the number of replicates does not have to be the same for each of the treatments. We could conduct the analyses comparing all pairs of means by doing many two-sample t-tests. However, this is strongly discouraged because it is not only inefficient, but the probability of making at least one Type I error by conducting these multiple comparisons increases as more and more tests are conducted. It is recommended, therefore, that an overall comparison of means be conducted first, and subsequent pairwise comparisons can be made if the overall comparisons indicates that at least one of the means is different from the others.
Capturing student feedback literacy using reflective logs
Published in European Journal of Engineering Education, 2023
Kurt Coppens, Lynn Van den Broeck, Naomi Winstone, Greet Langie
A variety of reflection levels was observed during the coding process. Therefore, after the coding, the level of reflection was additionally assessed for each reflective log using the assessment scheme in Table 2. It is based on the four-category scheme by Kember et al. (2008), which was developed to evaluate the level of reflection in students’ written work, and was adapted in the present study to fit the feedback reflections. Multiple pairwise Fisher's Exact Tests were performed to determine if there was an association between the reflection level and the presence of the different feedback literacy characteristics (categories ‘none’, ‘basic’ or ‘advanced’). A Bonferroni correction was applied for counteracting the multiple comparisons.
Thoracic limb activity of Simocephalus vetulus and its descendants is shaped by the combination of delafloxacin and calcium
Published in Human and Ecological Risk Assessment: An International Journal, 2023
Tan-Duc Nguyen, Tomoaki Itayama, Norio Iwami, Kazuya Shimizu, Thanh-Son Dao, Thanh Luu Pham, Hideaki Maseda
Toxicity testing often uses several levels of stressors; for example, this study used eight treatments, corresponding to 28 possible simultaneous hypothesis tests. The multiple comparison problem leads to the false positive error, where the null hypothesis is rejected, even though it is true (Lee and Lee 2018). The cause is the inflated probability of significant outcomes by chance (Streiner 2015). To address this issue in statistics, several methods have been developed to adjust the raw p-value; however, all known methods inevitably decrease the ability to detect true positives, particularly when a large number of hypotheses are tested simultaneously, which is considered a tradeoff mechanism for controlling the issue of multiplicity (see Menyhart 2021). Further, the null hypothesis-rejecting power among methods is different (Chen et al. 2017). Accordingly, the most conservative approach is the Bonferroni adjustment, while the Benjamini-Hochberg adjustment is considered the most powerful method (see Chen et al. 2017). Therefore, the choice of the p-value adjustment methods can literally shape scientific findings. In addition to the multiplicity issue, conducting hypothesis testing with a small sample size, due to specific labor and experimental conditions, can lead to p-values > 0.05 (i.e., no significant difference), even when the effect size (i.e., mean difference) is considerably large (see Kim and Bang 2016).
Associations between grass pollen exposures in utero and in early life with food allergy in 12-month-old infants
Published in International Journal of Environmental Health Research, 2022
Nugroho Harry Susanto, Adrian J Lowe, Agus Salim, Jennifer J. Koplin, Mimi L. K. Tang, Noor H. A. Suaini, Anne-Louise Ponsonby, Katrina J. Allen, Shyamali C. Dharmage, Bircan Erbas
A potential limitation is that we did not have pollen measurement outside the predefined pollen season. Although the start and end timing slightly vary, grass pollen season seems to occur during October to early-January with about 90% of the annual pollen load occurring during this period. Studies have shown that low or clinically irrelevant levels outside this period (Ong et al. 1995a; Beggs et al. 2015). Therefore, the impacts on the outcome are likely to be negligible. We had no data on time spent outdoors; it is possible that mothers reporting hay fever symptoms may remain indoors during high pollen days. Therefore, the effects are likely to be stronger than what s reported. We also only used mother’s report to define mother’s history of hay fever and food allergy, which might have led to misclassification. This misclassification might contribute to decreasing the effect of mother’s history of hay fever and food allergy to null. We do not have age at exposure to first respiratory virus season as this is a plausible confounder. Although this study is exploratory, it is important to consider the issue of multiple comparisons. Several analyses have been undertaken and there seems to be a consistent finding for grass pollen exposure in utero and some food allergy outcomes.