Statistical P values do not dominate scientific research / 中华预防医学杂志

Statistical P value and its threshold have been controversial worldwide for a while. Recent heated debate was triggered by two practical issues: unexplainable high false positive rate in biomedical research, and global misunderstood of "statistical significance" in scientific community. Thus, part of scientists suggests applying more stringent significance level (from 0.05 to 0.005), or even giving up the use of significance level. We believe that they are throwing the baby out with the bath water. These suggestions will not contribute to any improvement of this unfavorable situation but will lead the scientific decision-making to a more difficult and subjective corner. Scientists should use statistical P value and threshold only if they correctly understand the soul of statistics-uncertainty. Statistical significance is neither sole nor dominant criterion to measure the scientific value, but an honest assistant. Scientific decision-making should initiate from the scientific experimental design, followed by rigorous implementation and transparent analysis, and synthesize a variety of information to reach a tenable conclusion.

Essentials of Statistics in Dental Research: Part 1.

Dental professionals review articles in scientific dental journals to be informed about progress in the field and to remain up to date in the dental literature. As a general rule, knowing the basic statistical concepts such as study design, hypothesis testing, confidence interval, significance level, effect size, and sample size and power can help readers to better understand the whole concept of an article. Reviews of statistical methodology in published papers of well-established dental journals concluded that the majority of the papers contained statistical errors. Inappropriate use of statistical methods may lead to incorrect conclusions and a waste of valuable resources. The aim of this paper is to provide an overview on some implications of the statistical concepts and recommendations for avoiding statistical errors in dental research.

Endogamia e limite de seleção em populações sob seleção assistida por marcadores moleculares / Endogamy and its selection limit in populations under assisted selection by molecular markers

Foram utilizados diferentes níveis de significância genômica na seleção assistida por marcadores para estimar a endogamia média e o limite de seleção, assim como os valores fenotípicos, em características quantitativas de baixa, média e alta herdabilidade. Uma comparação entre os níveis de 1 por cento, 5 por cento, 10 por cento e 20 por cento foi realizada por meio do sistema computacional de simulação genética (GENESYS), utilizado para a simulação de três genomas (cada qual constituído de um único caráter de baixa, média ou alta herdabilidade), e das populações base e inicial. Os resultados indicaram superioridade dos níveis de significância de maior magnitude (10 por cento e 20 por cento) com relação aos valores fenotípicos, resultante de menor média endogâmica, além de menor limite de seleção ao longo das gerações sob seleção para estes níveis. Estes resultados foram observados para todas as três características, embora de forma mais expressiva para o caráter de baixa herdabilidade. Assim, apesar de os níveis de 1 por cento e 5 por cento apresentarem maior precisão na detecção de marcadores ligados a quantitative trait loci (QTL), eles conduzem a maiores médias endogâmicas e limite de seleção, propiciando ganhos fenotípicos menores.

Different levels of genomic significance were used in the selection assisted by molecular markers to estimate the medium endogamy and the selection limit, as well as the phenotypic value, for quantitative traits of low, medium, and high heritability. A comparison among the levels of genomic significance of 1 percent, 5 percent, 10 percent, and 20 percent was accomplished by a computer system of genetic simulation (GENESYS), used to simulate three genomes, each of them constituted by only one character of low, medium and high heritability, and to simulate the base and the initial populations. The results suggest superiority of higher significance levels (10 percent and 20 percent) for all phenotypic values, as a consequence of lower endogamy, and lower selection limit for low, medium, and high heritability traits, but in more expressive way for low heritability trait. Although the significant levels of 1 percent and 5 percent for molecular marker assisted selection showed a high precision in detecting markers related to a quantitative trait loci (QTLs), they lead to higher endogamy and selection limits, resulting in low phenotypic gains.

Efeitos por comparações e por experimento em interações de experimentos fatoriais / Comparisonwise and experimentwise effects in fatorial experiments interactions

Avaliar efeitos de interações é um dos principais objetivos dos experimentos fatoriais. Em experimentos com dois fatores A e B, com m e n níveis de cada fator, respectivamente, há m x n possíveis interações e (m-1)(n-1) graus de liberdade associados. Freqüentemente somente parte dessas interações contribui efetivamente para a Soma de Quadrados da Interação e pode ser interessante examiná-las. O uso de nível de significância menos rigoroso para interpretação do efeito da interação "por experimento", em relação às demais fontes de variação da análise de variância, pode captar efeitos importantes. Recomenda-se o uso de p = 0,25 para a interpretação do efeito da interação "por experimento", mantendo-se o usual p = 0,05 para efeitos "por comparações". Mesmo no caso de interações significativas, comparações selecionadas, em lugar de apenas "cortes", podem auxiliar a interpretação de interações complexas.

To evaluate interaction is one of the most important objectives of fatorial experiments. In experiments with two factors A and B, with m and n levels of each factor, there are m x n possible interactions and (m-1)(n-1) degrees of freedom associated. Frequently just a part of these interactions contributes effectively to the sum of squares of interaction and must be interesting to evaluate them. The use of less rigorous level of significance for interpretation of the interaction "for experiment", in relation to sources of variation of the variance analysis, may catch important effect. One recommends the use of p = 0.25 for the interpretation of the effect of the interaction "for experiment" remaining usual p = 0.05 for effect "for comparisons". Even in the case of significant interactions, selected comparisons, instead of "cuts" only, may assist the interpretation of complex interactions.