Peter E. Freeman
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Peter E. Freeman has a total of 87 co-authors. Here are the top co-authors by number of shared publications:
A. B. Lee
Department of Statistics, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213, USA
7 shared publicationsAnn B. Lee
Department of Statistics Carnegie Mellon University 5000 Forbes Avenue 15213 Pittsburgh PA USA
4 shared publicationsJ. A. Newman
4 shared publicationsChad M. Schafer
3 shared publicationsR. Izbicki
Department of Statistics, Federal University of São Carlos, 13565-905 São Carlos, Brazil
3 shared publications- Andreas Zezas
Physics Department & Institute of Theoretical and Computational Physics, University of Crete
2 shared publications
Latest Publications
- Open Access
- 7 August 2026
- Published by Taylor & Francis in Journal of Statistics and Data Science Education
- Vol. 34 (4) , 450-463
Modernizing How Students Compute Interval Estimates in Mathematical Statistics Courses
1. The term “mathematical statistics” denotes a calculus-based course sequence that covers concepts of probability and statistical inference, generally over the course of an academic year. This cou...
- 7 August 2026
- Published by Taylor & Francis in Journal of Statistics and Data Science Education
- Vol. 34 (4) , 450-463
Open AccessModernizing How Students Compute Interval Estimates in Mathematical Statistics Courses
1. The term “mathematical statistics” denotes a calculus-based course sequence that covers concepts of probability and statistical inference, generally over the course of an academic year. This cou...
- 14 December 2020
- Published by Taylor & Francis in The American Statistician
- Vol. 75 (4) , 433-444
Facilitating Authentic Practice for Early Undergraduate Statistics Students
In current curricula, authentic statistical practice generally only occurs in capstone projects undertaken by advanced undergraduate and Master’s students. We argue that deferring practice is a mistake: undergraduate students should achieve experience via repeated practice from their first years onward, to achieve heightened levels of confidence and competence prior to graduation. However, statistical practice is not a “one size fits all” enterprise: for instance, elements of a capstone experience, such as extensive data preprocessing, may be out of place in earlier practice settings due to less-experienced students’ relative lack of coding skill. We describe a course we have implemented at Carnegie Mellon University, currently open to second-year students, that provides a circumscribed opportunity for statistical practice that limits coding breadth, uses fully curated data, treats statistical learning models as “gray boxes” to be understood qualitatively, and provides open-ended semester-long projects that students pursue outside of class. We show how pre- and post-course assessment tests and retrospective surveys indicate clear gains in the students’ knowledge of, and attitudes toward, statistical practice. Given its clear benefits, we feel that statistics and data science programs should offer a course like the one we describe to all undergraduate students pursuing statistics and data science degrees.
- 14 December 2020
- Published by Taylor & Francis in The American Statistician
- Vol. 75 (4) , 433-444
Facilitating Authentic Practice for Early Undergraduate Statistics Students
In current curricula, authentic statistical practice generally only occurs in capstone projects undertaken by advanced undergraduate and Master’s students. We argue that deferring practice is a mistake: undergraduate students should achieve experience via repeated practice from their first years onward, to achieve heightened levels of confidence and competence prior to graduation. However, statistical practice is not a “one size fits all” enterprise: for instance, elements of a capstone experience, such as extensive data preprocessing, may be out of place in earlier practice settings due to less-experienced students’ relative lack of coding skill. We describe a course we have implemented at Carnegie Mellon University, currently open to second-year students, that provides a circumscribed opportunity for statistical practice that limits coding breadth, uses fully curated data, treats statistical learning models as “gray boxes” to be understood qualitatively, and provides open-ended semester-long projects that students pursue outside of class. We show how pre- and post-course assessment tests and retrospective surveys indicate clear gains in the students’ knowledge of, and attitudes toward, statistical practice. Given its clear benefits, we feel that statistics and data science programs should offer a course like the one we describe to all undergraduate students pursuing statistics and data science degrees.
- Open Access
- 18 July 2017
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 471 (3) , 3273-3282
Local two-sample testing: a new tool for analysing high-dimensional astronomical data
Modern surveys have provided the astronomical community with a flood of high-dimensional data, but analyses of these data often occur after their projection to lower dimensional spaces. In this work, we introduce a local two-sample hypothesis test framework that an analyst may directly apply to data in their native space. In this framework, the analyst defines two classes based on a response variable of interest (e.g. higher mass galaxies versus lower mass galaxies) and determines at arbitrary points in predictor space whether the local proportions of objects that belong to the two classes significantly differ from the global proportion. Our framework has a potential myriad of uses throughout astronomy; here, we demonstrate its efficacy by applying it to a sample of 2487 i-band-selected galaxies observed by the HST-ACS in four of the CANDELS programme fields. For each galaxy, we have seven morphological summary statistics along with an estimated stellar mass and star formation rate (SFR). We perform two studies: one in which we determine regions of the seven-dimensional space of morphological statistics where high-mass galaxies are significantly more numerous than low-mass galaxies, and vice versa, and another study where we use SFR in place of mass. We find that we are able to identify such regions, and show how high-mass/low-SFR regions are associated with concentrated and undisturbed galaxies, while galaxies in low-mass/high-SFR regions appear more extended and/or disturbed than their high-mass/low-SFR counterparts.
- 18 July 2017
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 471 (3) , 3273-3282
Open AccessLocal two-sample testing: a new tool for analysing high-dimensional astronomical data
Modern surveys have provided the astronomical community with a flood of high-dimensional data, but analyses of these data often occur after their projection to lower dimensional spaces. In this work, we introduce a local two-sample hypothesis test framework that an analyst may directly apply to data in their native space. In this framework, the analyst defines two classes based on a response variable of interest (e.g. higher mass galaxies versus lower mass galaxies) and determines at arbitrary points in predictor space whether the local proportions of objects that belong to the two classes significantly differ from the global proportion. Our framework has a potential myriad of uses throughout astronomy; here, we demonstrate its efficacy by applying it to a sample of 2487 i-band-selected galaxies observed by the HST-ACS in four of the CANDELS programme fields. For each galaxy, we have seven morphological summary statistics along with an estimated stellar mass and star formation rate (SFR). We perform two studies: one in which we determine regions of the seven-dimensional space of morphological statistics where high-mass galaxies are significantly more numerous than low-mass galaxies, and vice versa, and another study where we use SFR in place of mass. We find that we are able to identify such regions, and show how high-mass/low-SFR regions are associated with concentrated and undisturbed galaxies, while galaxies in low-mass/high-SFR regions appear more extended and/or disturbed than their high-mass/low-SFR counterparts.
- Open Access
Local Two-Sample Testing: A New Tool for Analysing High-Dimensional Astronomical Data
Modern surveys have provided the astronomical community with a flood of high-dimensional data, but analyses of these data often occur after their projection to lower-dimensional spaces. In this work, we introduce a local two-sample hypothesis test framework that an analyst may directly apply to data in their native space. In this framework, the analyst defines two classes based on a response variable of interest (e.g. higher-mass galaxies versus lower-mass galaxies) and determines at arbitrary points in predictor space whether the local proportions of objects that belong to the two classes significantly differs from the global proportion. Our framework has a potential myriad of uses throughout astronomy; here, we demonstrate its efficacy by applying it to a sample of 2487 i-band-selected galaxies observed by the HST ACS in four of the CANDELS program fields. For each galaxy, we have seven morphological summary statistics along with an estimated stellar mass and star-formation rate. We perform two studies: one in which we determine regions of the seven-dimensional space of morphological statistics where high-mass galaxies are significantly more numerous than low-mass galaxies, and vice-versa, and another study where we use SFR in place of mass. We find that we are able to identify such regions, and show how high-mass/low-SFR regions are associated with concentrated and undisturbed galaxies while galaxies in low-mass/high-SFR regions appear more extended and/or disturbed than their high-mass/low-SFR counterparts.
Open AccessLocal Two-Sample Testing: A New Tool for Analysing High-Dimensional Astronomical Data
Modern surveys have provided the astronomical community with a flood of high-dimensional data, but analyses of these data often occur after their projection to lower-dimensional spaces. In this work, we introduce a local two-sample hypothesis test framework that an analyst may directly apply to data in their native space. In this framework, the analyst defines two classes based on a response variable of interest (e.g. higher-mass galaxies versus lower-mass galaxies) and determines at arbitrary points in predictor space whether the local proportions of objects that belong to the two classes significantly differs from the global proportion. Our framework has a potential myriad of uses throughout astronomy; here, we demonstrate its efficacy by applying it to a sample of 2487 i-band-selected galaxies observed by the HST ACS in four of the CANDELS program fields. For each galaxy, we have seven morphological summary statistics along with an estimated stellar mass and star-formation rate. We perform two studies: one in which we determine regions of the seven-dimensional space of morphological statistics where high-mass galaxies are significantly more numerous than low-mass galaxies, and vice-versa, and another study where we use SFR in place of mass. We find that we are able to identify such regions, and show how high-mass/low-SFR regions are associated with concentrated and undisturbed galaxies while galaxies in low-mass/high-SFR regions appear more extended and/or disturbed than their high-mass/low-SFR counterparts.
- 29 March 2017
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 468 (4) , 4556-4565
A unified framework for constructing, tuning and assessing photometric redshift density estimates in a selection bias setting
Photometric redshift estimation is an indispensable tool of precision cosmology. One problem that plagues the use of this tool in the era of large-scale sky surveys is that the bright galaxies that are selected for spectroscopic observation do not have properties that match those of (far more numerous) dimmer galaxies; thus, ill-designed empirical methods that produce accurate and precise redshift estimates for the former generally will not produce good estimates for the latter. In this paper, we provide a principled framework for generating conditional density estimates (i.e. photometric redshift PDFs) that takes into account selection bias and the covariate shift that this bias induces. We base our approach on the assumption that the probability that astronomers label a galaxy (i.e. determine its spectroscopic redshift) depends only on its measured (photometric and perhaps other) properties |$\boldsymbol {x}$| and not on its true redshift. With this assumption, we can explicitly write down risk functions that allow us to both tune and compare methods for estimating importance weights (i.e. the ratio of densities of unlabelled and labelled galaxies for different values of |$\boldsymbol {x}$|) and conditional densities. We also provide a method for combining multiple conditional density estimates for the same galaxy into a single estimate with better properties. We apply our risk functions to an analysis of ≈106 galaxies, mostly observed by Sloan Digital Sky Survey, and demonstrate through multiple diagnostic tests that our method achieves good conditional density estimates for the unlabelled galaxies.
- 29 March 2017
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 468 (4) , 4556-4565
A unified framework for constructing, tuning and assessing photometric redshift density estimates in a selection bias setting
Photometric redshift estimation is an indispensable tool of precision cosmology. One problem that plagues the use of this tool in the era of large-scale sky surveys is that the bright galaxies that are selected for spectroscopic observation do not have properties that match those of (far more numerous) dimmer galaxies; thus, ill-designed empirical methods that produce accurate and precise redshift estimates for the former generally will not produce good estimates for the latter. In this paper, we provide a principled framework for generating conditional density estimates (i.e. photometric redshift PDFs) that takes into account selection bias and the covariate shift that this bias induces. We base our approach on the assumption that the probability that astronomers label a galaxy (i.e. determine its spectroscopic redshift) depends only on its measured (photometric and perhaps other) properties |$\boldsymbol {x}$| and not on its true redshift. With this assumption, we can explicitly write down risk functions that allow us to both tune and compare methods for estimating importance weights (i.e. the ratio of densities of unlabelled and labelled galaxies for different values of |$\boldsymbol {x}$|) and conditional densities. We also provide a method for combining multiple conditional density estimates for the same galaxy into a single estimate with better properties. We apply our risk functions to an analysis of ≈106 galaxies, mostly observed by Sloan Digital Sky Survey, and demonstrate through multiple diagnostic tests that our method achieves good conditional density estimates for the unlabelled galaxies.
- Open Access
A Unified Framework for Constructing, Tuning and Assessing Photometric Redshift Density Estimates in a Selection Bias Setting
Photometric redshift estimation is an indispensable tool of precision cosmology. One problem that plagues the use of this tool in the era of large-scale sky surveys is that the bright galaxies that are selected for spectroscopic observation do not have properties that match those of (far more numerous) dimmer galaxies; thus, ill-designed empirical methods that produce accurate and precise redshift estimates for the former generally will not produce good estimates for the latter. In this paper, we provide a principled framework for generating conditional density estimates (i.e. photometric redshift PDFs) that takes into account selection bias and the covariate shift that this bias induces. We base our approach on the assumption that the probability that astronomers label a galaxy (i.e. determine its spectroscopic redshift) depends only on its measured (photometric and perhaps other) properties x and not on its true redshift. With this assumption, we can explicitly write down risk functions that allow us to both tune and compare methods for estimating importance weights (i.e. the ratio of densities of unlabeled and labeled galaxies for different values of x) and conditional densities. We also provide a method for combining multiple conditional density estimates for the same galaxy into a single estimate with better properties. We apply our risk functions to an analysis of approximately one million galaxies, mostly observed by SDSS, and demonstrate through multiple diagnostic tests that our method achieves good conditional density estimates for the unlabeled galaxies.
Open AccessA Unified Framework for Constructing, Tuning and Assessing Photometric Redshift Density Estimates in a Selection Bias Setting
Photometric redshift estimation is an indispensable tool of precision cosmology. One problem that plagues the use of this tool in the era of large-scale sky surveys is that the bright galaxies that are selected for spectroscopic observation do not have properties that match those of (far more numerous) dimmer galaxies; thus, ill-designed empirical methods that produce accurate and precise redshift estimates for the former generally will not produce good estimates for the latter. In this paper, we provide a principled framework for generating conditional density estimates (i.e. photometric redshift PDFs) that takes into account selection bias and the covariate shift that this bias induces. We base our approach on the assumption that the probability that astronomers label a galaxy (i.e. determine its spectroscopic redshift) depends only on its measured (photometric and perhaps other) properties x and not on its true redshift. With this assumption, we can explicitly write down risk functions that allow us to both tune and compare methods for estimating importance weights (i.e. the ratio of densities of unlabeled and labeled galaxies for different values of x) and conditional densities. We also provide a method for combining multiple conditional density estimates for the same galaxy into a single estimate with better properties. We apply our risk functions to an analysis of approximately one million galaxies, mostly observed by SDSS, and demonstrate through multiple diagnostic tests that our method achieves good conditional density estimates for the unlabeled galaxies.
- Open Access
Photo-z Estimation: An Example of Nonparametric Conditional Density Estimation under Selection Bias
Redshift is a key quantity for inferring cosmological model parameters. In photometric redshift estimation, cosmologists use the coarse data collected from the vast majority of galaxies to predict the redshift of individual galaxies. To properly quantify the uncertainty in the predictions, however, one needs to go beyond standard regression and instead estimate the full conditional density f(z|x) of a galaxy's redshift z given its photometric covariates x. The problem is further complicated by selection bias: usually only the rarest and brightest galaxies have known redshifts, and these galaxies have characteristics and measured covariates that do not necessarily match those of more numerous and dimmer galaxies of unknown redshift. Unfortunately, there is not much research on how to best estimate complex multivariate densities in such settings. Here we describe a general framework for properly constructing and assessing nonparametric conditional density estimators under selection bias, and for combining two or more estimators for optimal performance. We propose new improved photo-z estimators and illus- trate our methods on data from the Sloan Data Sky Survey and an application to galaxy-galaxy lensing. Although our main application is photo-z estimation, our methods are relevant to any high-dimensional regression setting with complicated asymmetric and multimodal distributions in the response variable.
Open AccessPhoto-z Estimation: An Example of Nonparametric Conditional Density Estimation under Selection Bias
Redshift is a key quantity for inferring cosmological model parameters. In photometric redshift estimation, cosmologists use the coarse data collected from the vast majority of galaxies to predict the redshift of individual galaxies. To properly quantify the uncertainty in the predictions, however, one needs to go beyond standard regression and instead estimate the full conditional density f(z|x) of a galaxy's redshift z given its photometric covariates x. The problem is further complicated by selection bias: usually only the rarest and brightest galaxies have known redshifts, and these galaxies have characteristics and measured covariates that do not necessarily match those of more numerous and dimmer galaxies of unknown redshift. Unfortunately, there is not much research on how to best estimate complex multivariate densities in such settings. Here we describe a general framework for properly constructing and assessing nonparametric conditional density estimators under selection bias, and for combining two or more estimators for optimal performance. We propose new improved photo-z estimators and illus- trate our methods on data from the Sloan Data Sky Survey and an application to galaxy-galaxy lensing. Although our main application is photo-z estimation, our methods are relevant to any high-dimensional regression setting with complicated asymmetric and multimodal distributions in the response variable.
- 2 February 2016
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 458 (1) , 963-987
Beyond spheroids and discs: classifications of CANDELS galaxy structure at 1.4 <z< 2 via principal component analysis
Important but rare and subtle processes driving galaxy morphology and star formation may be missed by traditional spiral, elliptical, irregular or Sérsic bulge/disc classifications. To overcome this limitation, we use a principal component analysis (PCA) of non-parametric morphological indicators (concentration, asymmetry, Gini coefficient, M20, multimode, intensity and deviation) measured at rest-frame B band (corresponding to HST/WFC3 F125W at 1.4 <z < 2) to trace the natural distribution of massive (>1010 M⊙) galaxy morphologies. PCA quantifies the correlations between these morphological indicators and determines the relative importance of each. The first three principal components (PCs) capture ∼75 per cent of the variance inherent to our sample. We interpret the first PC as bulge strength, the second PC as dominated by concentration and the third PC as dominated by asymmetry. Both PC1 and PC2 correlate with the visual appearance of a central bulge and predict galaxy quiescence. PC1 is a better predictor of quenching than stellar mass, as good as other structural indicators (Sérsic-n or compactness). We divide the PCA results into groups using an agglomerative hierarchical clustering method. Unlike Sérsic, this classification scheme separates compact galaxies from larger, smooth protoelliptical systems, and star-forming disc-dominated clumpy galaxies from star-forming bulge-dominated asymmetric galaxies. Distinguishing between these galaxy structural types in a quantitative manner is an important step towards understanding the connections between morphology, galaxy assembly and star formation.
- 2 February 2016
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 458 (1) , 963-987
Beyond spheroids and discs: classifications of CANDELS galaxy structure at 1.4 <z< 2 via principal component analysis
Important but rare and subtle processes driving galaxy morphology and star formation may be missed by traditional spiral, elliptical, irregular or Sérsic bulge/disc classifications. To overcome this limitation, we use a principal component analysis (PCA) of non-parametric morphological indicators (concentration, asymmetry, Gini coefficient, M20, multimode, intensity and deviation) measured at rest-frame B band (corresponding to HST/WFC3 F125W at 1.4 <z < 2) to trace the natural distribution of massive (>1010 M⊙) galaxy morphologies. PCA quantifies the correlations between these morphological indicators and determines the relative importance of each. The first three principal components (PCs) capture ∼75 per cent of the variance inherent to our sample. We interpret the first PC as bulge strength, the second PC as dominated by concentration and the third PC as dominated by asymmetry. Both PC1 and PC2 correlate with the visual appearance of a central bulge and predict galaxy quiescence. PC1 is a better predictor of quenching than stellar mass, as good as other structural indicators (Sérsic-n or compactness). We divide the PCA results into groups using an agglomerative hierarchical clustering method. Unlike Sérsic, this classification scheme separates compact galaxies from larger, smooth protoelliptical systems, and star-forming disc-dominated clumpy galaxies from star-forming bulge-dominated asymmetric galaxies. Distinguishing between these galaxy structural types in a quantitative manner is an important step towards understanding the connections between morphology, galaxy assembly and star formation.
- Open Access
- 9 April 2014
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 440 (3) , 2599-2609
Non-parametric 3D map of the intergalactic medium using the Lyman-alpha forest
Visualizing the high-redshift Universe is difficult due to the dearth of available data; however, the Lyman-alpha forest provides a means to map the intergalactic medium at redshifts not accessible to large galaxy surveys. Large-scale structure surveys, such as the Baryon Oscillation Spectroscopic Survey (BOSS), have collected quasar (QSO) spectra that enable the reconstruction of H i density fluctuations. The data fall on a collection of lines defined by the lines of sight (LOS) of the QSO, and a major issue with producing a 3D reconstruction is determining how to model the regions between the LOS. We present a method that produces a 3D map of this relatively uncharted portion of the Universe by employing local polynomial smoothing, a non-parametric methodology. The performance of the method is analysed on simulated data that mimics the varying number of LOS expected in real data, and then is applied to a sample region selected from BOSS. Evaluation of the reconstruction is assessed by considering various features of the predicted 3D maps including visual comparison of slices, probability density functions (PDFs), counts of local minima and maxima, and standardized correlation functions. This 3D reconstruction allows for an initial investigation of the topology of this portion of the Universe using persistent homology.
- 9 April 2014
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 440 (3) , 2599-2609
Open AccessNon-parametric 3D map of the intergalactic medium using the Lyman-alpha forest
Visualizing the high-redshift Universe is difficult due to the dearth of available data; however, the Lyman-alpha forest provides a means to map the intergalactic medium at redshifts not accessible to large galaxy surveys. Large-scale structure surveys, such as the Baryon Oscillation Spectroscopic Survey (BOSS), have collected quasar (QSO) spectra that enable the reconstruction of H i density fluctuations. The data fall on a collection of lines defined by the lines of sight (LOS) of the QSO, and a major issue with producing a 3D reconstruction is determining how to model the regions between the LOS. We present a method that produces a 3D map of this relatively uncharted portion of the Universe by employing local polynomial smoothing, a non-parametric methodology. The performance of the method is analysed on simulated data that mimics the varying number of LOS expected in real data, and then is applied to a sample region selected from BOSS. Evaluation of the reconstruction is assessed by considering various features of the predicted 3D maps including visual comparison of slices, probability density functions (PDFs), counts of local minima and maxima, and standardized correlation functions. This 3D reconstruction allows for an initial investigation of the topology of this portion of the Universe using persistent homology.