E. Michael Nussbaum
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
E. Michael Nussbaum has a total of 36 co-authors. Here are the top co-authors by number of shared publications:
- Michael S. Van Winkle
Department of Educational Psychology, Leadership, & Higher Education, University of Nevada, Las Vegas, Las Vegas, NV 89154, USA
3 shared publications - Christina Nishiyama
Department of Educational Psychology, Leadership, and Higher Education, University of Nevada, Las Vegas, Las Vegas, USA
2 shared publications Gale M. Sinatra
University of Nevada
2 shared publications- Ian J. Dove
Department of Philosophy University of Nevada Las Vegas Nevada USA
2 shared publications - Jacqueline R. Cordova
University of Nevada, Las Vegas, College of Education, 4505 Maryland Parkway, Box 453003, Las Vegas, Nevada 89154-3003, USA
2 shared publications - Marissa C. Owens
Marissa C. Owens, doctoral graduate, University of Nevada, Las Vegas, NV, USA
2 shared publications
Latest Publications
- Open Access
- 15 October 2026
- Published by MDPI AG in Education Sciences
- Vol. 16 (10) , 1721
Tools for Promoting Conceptually Productive Discussion and Argumentation in Science Education
Teachers often have difficulty facilitating whole-class science discussions that are not teacher-centric. In this study, we explore the use of a planning/reflection form (PRF) to support these discussions in three middle-school science classrooms. The PRF was based on three instructional frameworks: Ian J. Dove and E. Michael Nussbaum’s Critical Questions Model of Argument Assessment, talk moves as described in Sarah Michaels and Catherine O’Connor’s Academically Productive Talk, and an elicit-explore-consensus instructional cycle. The frameworks also provided a basis for identifying both deficiencies and successes in the teachers’ discourse practices. Data included recordings of discussions and teacher reflection meetings, analyzed for thematic discursive patterns. We found the curriculum (Amplify Science) incorporated some critical questions for evaluating arguments, but provided limited scaffolding for facilitating whole-class discussions or probing student reasoning. Teachers viewed the PRF as helping them focus on student contributions and formative assessment data. One teacher also implemented a student science seminar led by student moderators. This was a form of whole-class discussion that was not teacher-centric but still conceptually productive. We argue that current conceptions of argumentation in science education should be broadened to place greater emphasis on learning to participate in conceptually productive discussions.
- 15 October 2026
- Published by MDPI AG in Education Sciences
- Vol. 16 (10) , 1721
Open AccessTools for Promoting Conceptually Productive Discussion and Argumentation in Science Education
Teachers often have difficulty facilitating whole-class science discussions that are not teacher-centric. In this study, we explore the use of a planning/reflection form (PRF) to support these discussions in three middle-school science classrooms. The PRF was based on three instructional frameworks: Ian J. Dove and E. Michael Nussbaum’s Critical Questions Model of Argument Assessment, talk moves as described in Sarah Michaels and Catherine O’Connor’s Academically Productive Talk, and an elicit-explore-consensus instructional cycle. The frameworks also provided a basis for identifying both deficiencies and successes in the teachers’ discourse practices. Data included recordings of discussions and teacher reflection meetings, analyzed for thematic discursive patterns. We found the curriculum (Amplify Science) incorporated some critical questions for evaluating arguments, but provided limited scaffolding for facilitating whole-class discussions or probing student reasoning. Teachers viewed the PRF as helping them focus on student contributions and formative assessment data. One teacher also implemented a student science seminar led by student moderators. This was a form of whole-class discussion that was not teacher-centric but still conceptually productive. We argue that current conceptions of argumentation in science education should be broadened to place greater emphasis on learning to participate in conceptually productive discussions.
- 11 May 2024
- Published by Wiley in Science Education
- Vol. 108 (5) , 1420-1447
Extending science instruction beyond the CER: Use of critical questions in the argumentation of middle school science students
Critiquing arguments is important for K‐12 science students to learn but not emphasized by the predominant claim‐evidence‐reasoning (CER) argumentation model. Drawing on the work of Yu and Zenker (2020), and Dove and Nussbaum (2018), we developed a tool for supplementing CER with critical questions (CQs) from philosophy that cover most, if not all, the logical dimensions of argument critique. Six middle school science teachers designed lessons involving argumentation, including the use of CQs. We assessed the effects on student self‐efficacy for engaging in argument critique, teacher self‐efficacy for using argument pedagogy, and teachers’ perceptions of the value of CQs. Qualitative data included teacher interviews, lesson transcripts, and student work samples. Quantitative data included surveys of student self‐efficacy administered at the beginning and end of the school year. There was evidence suggesting an increase over time in students’ confidence for engaging in argument critique and teachers’ confidence with argument pedagogy. However, only four of the six teachers were confident and skilled enough to include CQs in their lessons. Those who did use CQs tended to perceive them as providing a helpful structure for critique, prompts for deeper thinking, and a tool for fostering critical classroom norms. Discussion of CQs may have benefitted students’ writing by promoting peer critique and encouragement to elaborate. Overall, CQs afford students with a framework for judging argument strength. Scientific argumentation involving CQs provides a more contemporary philosophical basis for scientific argumentation than CER or the Toulmin model as it emphasizes the critical and dialogic nature of science.
- 11 May 2024
- Published by Wiley in Science Education
- Vol. 108 (5) , 1420-1447
Extending science instruction beyond the CER: Use of critical questions in the argumentation of middle school science students
Critiquing arguments is important for K‐12 science students to learn but not emphasized by the predominant claim‐evidence‐reasoning (CER) argumentation model. Drawing on the work of Yu and Zenker (2020), and Dove and Nussbaum (2018), we developed a tool for supplementing CER with critical questions (CQs) from philosophy that cover most, if not all, the logical dimensions of argument critique. Six middle school science teachers designed lessons involving argumentation, including the use of CQs. We assessed the effects on student self‐efficacy for engaging in argument critique, teacher self‐efficacy for using argument pedagogy, and teachers’ perceptions of the value of CQs. Qualitative data included teacher interviews, lesson transcripts, and student work samples. Quantitative data included surveys of student self‐efficacy administered at the beginning and end of the school year. There was evidence suggesting an increase over time in students’ confidence for engaging in argument critique and teachers’ confidence with argument pedagogy. However, only four of the six teachers were confident and skilled enough to include CQs in their lessons. Those who did use CQs tended to perceive them as providing a helpful structure for critique, prompts for deeper thinking, and a tool for fostering critical classroom norms. Discussion of CQs may have benefitted students’ writing by promoting peer critique and encouragement to elaborate. Overall, CQs afford students with a framework for judging argument strength. Scientific argumentation involving CQs provides a more contemporary philosophical basis for scientific argumentation than CER or the Toulmin model as it emphasizes the critical and dialogic nature of science.
- 26 March 2024
- Published by Taylor & Francis
- p. 81-98
Contingency Tables
This chapter examines contingency tables and how to conduct a chi-square test for statistical independence between two nominal variables in such tables. It also examines associated effect size measures. The structure of 2 × 2 contingency tables is first explained, along with the concept of conditional probability. The logic of the chi-squared test of statistical independence between two nominal variables is then explored and the underlying logic elucidated. It is noted that If the chi-squared test is significant, it should be reported with an effect size measure so that practical significance can be judged. Measures that are addressed include difference in proportions, Cohen’s H, odds ratios, phi correlations, and Cramer’s V. The first two of these assume that one variable is random and the other fixed. The other measures are symmetric and do not require designating a dependent and independent variable. The chapter also provides steps in IBM SPSS, R, and Excel for calculating these various statistics.
- 26 March 2024
- Published by Taylor & Francis
- p. 81-98
Contingency Tables
This chapter examines contingency tables and how to conduct a chi-square test for statistical independence between two nominal variables in such tables. It also examines associated effect size measures. The structure of 2 × 2 contingency tables is first explained, along with the concept of conditional probability. The logic of the chi-squared test of statistical independence between two nominal variables is then explored and the underlying logic elucidated. It is noted that If the chi-squared test is significant, it should be reported with an effect size measure so that practical significance can be judged. Measures that are addressed include difference in proportions, Cohen’s H, odds ratios, phi correlations, and Cramer’s V. The first two of these assume that one variable is random and the other fixed. The other measures are symmetric and do not require designating a dependent and independent variable. The chapter also provides steps in IBM SPSS, R, and Excel for calculating these various statistics.
- 26 March 2024
- Published by Taylor & Francis
- p. 129-173
Basic Nonparametric Tests for Ordinal Data
The chapter first discusses situations in which nonparametric techniques should be considered: Specifically, when there are ordinal or bounded variables, small samples, or when distributions are skewed or kurtotic. Techniques for analyzing the relationship of two ordinal variables are then presented, specifically Spearman’s rho and Kendall’s tau. (The latter can also be used with ordered categories.) Somers’ d and the Jonckheere-Terpstra test are also presented. The chapter then examines techniques for comparing two groups, specifically the Median Test and the Wilcoxon-Mann-Whitney Test. For the latter, the null hypothesis can be framed either in terms of identical distribution functions or identical relative effects (also known as relative treatment effects). Relative effects involve assessing the probability that a randomly selected case from one group will rank higher than one for the other. They can be used as an effect size measure, as can the Hodges-Lehmann estimator when the dependent variable is metric. The chapter then describes the Behrens-Fisher problem, which relates to unequal variances, and procedures for testing whether the variances of two groups are equal. These tests of dispersion (or scale) include the Mood test and Fligner-Killeen test.
- 26 March 2024
- Published by Taylor & Francis
- p. 129-173
Basic Nonparametric Tests for Ordinal Data
The chapter first discusses situations in which nonparametric techniques should be considered: Specifically, when there are ordinal or bounded variables, small samples, or when distributions are skewed or kurtotic. Techniques for analyzing the relationship of two ordinal variables are then presented, specifically Spearman’s rho and Kendall’s tau. (The latter can also be used with ordered categories.) Somers’ d and the Jonckheere-Terpstra test are also presented. The chapter then examines techniques for comparing two groups, specifically the Median Test and the Wilcoxon-Mann-Whitney Test. For the latter, the null hypothesis can be framed either in terms of identical distribution functions or identical relative effects (also known as relative treatment effects). Relative effects involve assessing the probability that a randomly selected case from one group will rank higher than one for the other. They can be used as an effect size measure, as can the Hodges-Lehmann estimator when the dependent variable is metric. The chapter then describes the Behrens-Fisher problem, which relates to unequal variances, and procedures for testing whether the variances of two groups are equal. These tests of dispersion (or scale) include the Mood test and Fligner-Killeen test.
- 26 March 2024
- Published by Taylor & Francis
- p. 297-347
Binary Logistic Regression
This chapter addresses logistic regression with a binary dependent variable. The logistic curve is transformed into a linear one by using log odds as the dependent variable in the linear equation. Logistic regression employs maximum likelihood estimation (MLE) to estimate parameters; the estimated likelihoods of models with and without a specific variable can then be used to perform a likelihood ratio (LR) test of statistical significance. Other methods of testing significance include the Wald and score tests. Various fit statistics are described, including the deviance, Pearson’s chi-square, Pseudo-R2, the Akaike information criterion (AIC), the Bayesian information criterion (BIC), the Hosmer-Lemeshow statistic, and percentage of cases classified correctly. Some recommended steps are outlined for building and refining logistic models. Procedures for checking for outliers, nonlinearity in the logits, alternative link functions, and overdispersion with grouped data are addressed.
- 26 March 2024
- Published by Taylor & Francis
- p. 297-347
Binary Logistic Regression
This chapter addresses logistic regression with a binary dependent variable. The logistic curve is transformed into a linear one by using log odds as the dependent variable in the linear equation. Logistic regression employs maximum likelihood estimation (MLE) to estimate parameters; the estimated likelihoods of models with and without a specific variable can then be used to perform a likelihood ratio (LR) test of statistical significance. Other methods of testing significance include the Wald and score tests. Various fit statistics are described, including the deviance, Pearson’s chi-square, Pseudo-R2, the Akaike information criterion (AIC), the Bayesian information criterion (BIC), the Hosmer-Lemeshow statistic, and percentage of cases classified correctly. Some recommended steps are outlined for building and refining logistic models. Procedures for checking for outliers, nonlinearity in the logits, alternative link functions, and overdispersion with grouped data are addressed.
- 26 March 2024
- Published by Taylor & Francis
- p. 349-386
Multinomial Logistic, Ordinal, and Poisson Regression
This chapter considers regression techniques where the dependent variable consists of multiple categories or counts. Multinomial logistic regression is used when the dependent variable consists of unordered categories; it is based on the multinomial distribution. When the dependent variable consists of ordered categories, ordinal regression can be used. Ordinal regression is known by a number of other names, including cumulative logistic regression, cumulative logits, the Polytomous Universal Model (PLUM), and the proportional odds model. The chapter discusses the underlying model, tests of the parallel lines assumption, and remedies for when this assumption is violated. These include using alternative link functions, such as the complementary log-log (cloglog) link and the negative log-log link. Poisson regression is used when the dependent variable consists of counts. The chapter compares and contrasts the Poisson distribution with the binomial, and then presents the Poisson Generalized Linear Model (GLM). The use of an offset in Poisson regression is explained. The Poisson GLM assumes that the variance of the dependent variable is equal to the mean. Violations of this assumption produces overdispersion or underdispersion. The chapter presents negative binomial regression as a technique for assessing and addressing overdispersion.
- 26 March 2024
- Published by Taylor & Francis
- p. 349-386
Multinomial Logistic, Ordinal, and Poisson Regression
This chapter considers regression techniques where the dependent variable consists of multiple categories or counts. Multinomial logistic regression is used when the dependent variable consists of unordered categories; it is based on the multinomial distribution. When the dependent variable consists of ordered categories, ordinal regression can be used. Ordinal regression is known by a number of other names, including cumulative logistic regression, cumulative logits, the Polytomous Universal Model (PLUM), and the proportional odds model. The chapter discusses the underlying model, tests of the parallel lines assumption, and remedies for when this assumption is violated. These include using alternative link functions, such as the complementary log-log (cloglog) link and the negative log-log link. Poisson regression is used when the dependent variable consists of counts. The chapter compares and contrasts the Poisson distribution with the binomial, and then presents the Poisson Generalized Linear Model (GLM). The use of an offset in Poisson regression is explained. The Poisson GLM assumes that the variance of the dependent variable is equal to the mean. Violations of this assumption produces overdispersion or underdispersion. The chapter presents negative binomial regression as a technique for assessing and addressing overdispersion.
- 26 March 2024
- Published by Taylor & Francis
- p. 175-225
Nonparametric Tests for Multiple Independent Samples
The logic behind comparing two independent samples nonparametrically—presented in Chapter 6—is extended in this chapter to multiple samples, specifically (1) the Kruskal-Wallis test for balanced designs and (2) the use of pseudo-ranks and unweighted relative effects for unbalanced designs. Techniques for pairwise comparison/pairwise contrasts are then discussed. The chapter explains why the use of multi-distributional relative effects, where comparisons are made against the average distribution function, are superior to pairwise relative effects, because the latter can result in intransitive decisions. Another issue considered is how the size of Bonferroni-adjustments can be reduced by making planned comparisons or eliminated entirely by using multiple contrast tests and/or simultaneous confidence intervals. Bonferroni-adjustments are also unneeded when only three groups are involved because of the closed testing principle. The chapter next presents the Hettmansperger-Norton test for analyzing patterned alternatives, for example when treatments are ordered. The chapter also considers nonparametric analyses of more complex factorial designs involving interactions, specifically those involving ANOVA-type statistics (ATS) or, for 2 × 2 designs, linear rank statistics.
- 26 March 2024
- Published by Taylor & Francis
- p. 175-225
Nonparametric Tests for Multiple Independent Samples
The logic behind comparing two independent samples nonparametrically—presented in Chapter 6—is extended in this chapter to multiple samples, specifically (1) the Kruskal-Wallis test for balanced designs and (2) the use of pseudo-ranks and unweighted relative effects for unbalanced designs. Techniques for pairwise comparison/pairwise contrasts are then discussed. The chapter explains why the use of multi-distributional relative effects, where comparisons are made against the average distribution function, are superior to pairwise relative effects, because the latter can result in intransitive decisions. Another issue considered is how the size of Bonferroni-adjustments can be reduced by making planned comparisons or eliminated entirely by using multiple contrast tests and/or simultaneous confidence intervals. Bonferroni-adjustments are also unneeded when only three groups are involved because of the closed testing principle. The chapter next presents the Hettmansperger-Norton test for analyzing patterned alternatives, for example when treatments are ordered. The chapter also considers nonparametric analyses of more complex factorial designs involving interactions, specifically those involving ANOVA-type statistics (ATS) or, for 2 × 2 designs, linear rank statistics.
- 26 March 2024
- Published by Taylor & Francis
- p. 417-441
General Estimating Equations
The general estimating equations (GEE) approach is used in lieu of generalized linear models when there is statistical dependence among observations, and specifically with longitudinal or clustered data. Quasi-likelihood estimation is used to estimate parameters. The chapter defines quasi-likelihoods and explains how these are more valid estimates of the probability of obtaining the observed data, given specific parameter estimates, than are likelihoods. Quasi-likelihoods are based, in part, on expected variances, which are based on “variance functions” expressing the relationship between expected variances and means. The variance function is used in combination with a correlational structure to estimate a covariance matrix for each individual or cluster. (Autoregressive, exchangeable, independent, and M-dependent structures are considered, and how the quasi-likelihood information criterion (QIC) can be used to compare the fit of these different structures.) Once beta parameters are estimated, standard errors are derived from the aggregate covariance matrix. A generalized score test can then be used to test for statistical significance. The chapter also covers the use of robust standard errors based on information sandwich estimators, how the GEE procedure addresses missing values, and how the procedure can be implemented in IBM SPSS and R.
- 26 March 2024
- Published by Taylor & Francis
- p. 417-441
General Estimating Equations
The general estimating equations (GEE) approach is used in lieu of generalized linear models when there is statistical dependence among observations, and specifically with longitudinal or clustered data. Quasi-likelihood estimation is used to estimate parameters. The chapter defines quasi-likelihoods and explains how these are more valid estimates of the probability of obtaining the observed data, given specific parameter estimates, than are likelihoods. Quasi-likelihoods are based, in part, on expected variances, which are based on “variance functions” expressing the relationship between expected variances and means. The variance function is used in combination with a correlational structure to estimate a covariance matrix for each individual or cluster. (Autoregressive, exchangeable, independent, and M-dependent structures are considered, and how the quasi-likelihood information criterion (QIC) can be used to compare the fit of these different structures.) Once beta parameters are estimated, standard errors are derived from the aggregate covariance matrix. A generalized score test can then be used to test for statistical significance. The chapter also covers the use of robust standard errors based on information sandwich estimators, how the GEE procedure addresses missing values, and how the procedure can be implemented in IBM SPSS and R.
- 26 March 2024
- Published by Taylor & Francis
- p. 481-485
Choosing the Best Statistical Technique
This chapter discusses the factors that need to be considered in choosing the best statistical technique for particular situations. Tables 15.1 and 15.2 present test matrices that take into account whether independent or related samples are used, the level of measurement of the independent and dependent variables (i.e., whether nominal, ordinal, or metric), the number of samples, whether samples are ordered, sample sizes, and whether potentially confounding variables need to be controlled. The matrices summarize most of the tests discussed in the book, both parametric and nonparametric ones. In deciding between two tests that might be otherwise appropriate, one needs to consider whether all the assumptions of a given test are likely to be met. Some tests, like those for the ANOVA-Type Statistic, are more robust than others because they make fewer assumptions. Some tests are also more statistically powerful than others, depending on the skew and kurtosis of the dependent variable in the population. These are all issues that need to be considered when choosing the best statistical technique.
- 26 March 2024
- Published by Taylor & Francis
- p. 481-485
Choosing the Best Statistical Technique
This chapter discusses the factors that need to be considered in choosing the best statistical technique for particular situations. Tables 15.1 and 15.2 present test matrices that take into account whether independent or related samples are used, the level of measurement of the independent and dependent variables (i.e., whether nominal, ordinal, or metric), the number of samples, whether samples are ordered, sample sizes, and whether potentially confounding variables need to be controlled. The matrices summarize most of the tests discussed in the book, both parametric and nonparametric ones. In deciding between two tests that might be otherwise appropriate, one needs to consider whether all the assumptions of a given test are likely to be met. Some tests, like those for the ANOVA-Type Statistic, are more robust than others because they make fewer assumptions. Some tests are also more statistically powerful than others, depending on the skew and kurtosis of the dependent variable in the population. These are all issues that need to be considered when choosing the best statistical technique.
- 26 March 2024
- Published by Taylor & Francis
- p. 99-127
Contingency Tables
This chapter examines tests for 2 × 2 contingency tables when it is inappropriate to use the chi-square test for statistical independence. These special situations include small samples, lack of statistical independence among observations, and the need to statistically control for confounding variables. The chapter specifically considers Fisher’s exact test (for small samples), the McNemar test (for two related samples), the Cochran test (when three or more time periods are involved), and the Mantel-Haenszel test (for controlling a third variable). Most of these tests are applicable only to 2 × 2 tables or 2 × 2 × k tables. To analyze larger tables, or to test statistical interactions, one can use loglinear models, which are briefly introduced in this chapter and described in more detail in Chapter 12. It is noted that the situations addressed in this chapter—small samples, related samples (for example pre-post designs), and controlling for extraneous variables—reflect general considerations that are relevant in choosing any statistical test, not just those involving nominal variables.
- 26 March 2024
- Published by Taylor & Francis
- p. 99-127
Contingency Tables
This chapter examines tests for 2 × 2 contingency tables when it is inappropriate to use the chi-square test for statistical independence. These special situations include small samples, lack of statistical independence among observations, and the need to statistically control for confounding variables. The chapter specifically considers Fisher’s exact test (for small samples), the McNemar test (for two related samples), the Cochran test (when three or more time periods are involved), and the Mantel-Haenszel test (for controlling a third variable). Most of these tests are applicable only to 2 × 2 tables or 2 × 2 × k tables. To analyze larger tables, or to test statistical interactions, one can use loglinear models, which are briefly introduced in this chapter and described in more detail in Chapter 12. It is noted that the situations addressed in this chapter—small samples, related samples (for example pre-post designs), and controlling for extraneous variables—reflect general considerations that are relevant in choosing any statistical test, not just those involving nominal variables.