The AstroStat Slog

Astronomers rely on scatter plots to illustrate correlations and trends among many pairs of variables more than any scientists^[1]. Pages of scatter plots with regression lines are often found from which the slope of regression line and errors bars are indicators of degrees of correlation. Sometimes, too many of such scatter plots makes me think that, overall, resources for drawing nice scatter plots and papers where those plots are printed are wasted. Why not just compute correlation coefficients and its error and publicize the processed data for computing correlations, not the full data, so that others can verify the computation results for the sake of validation? A couple of scatter plots are fine but when I see dozens of them, I lost my focus. This is another cultural difference. Continue reading ‘Scatter plots and ANCOVA’ »

1. This is not an assuring absolute statement but a personal impression after reading articles of various fields in addition to astronomy. My readings of other fields tell that many rely on correlation statistics but less scatter plots by adding straight lines going through data sets for the purpose of imposing relationships within variable pairs[↩]

My first impression from the 212th AAS meeting is that it’s planned for preparing IYA 2009 and many talks are about current and future project reviews and strategies to reach public (People kept saying to me that winter meetings are more grand with expanded topics). I cannot say I understand everything (If someone says no astronomers understand everything, I’ll be relieved) but thanks to the theme of the meeting, I was intelligently entertained enough in many respects. The downside of this intellectual stimulus is growing doubts. One of those doubts was regression analysis in astronomy. Continue reading ‘my first AAS. I. Regression’ »

One of the papers from arxiv/astro-ph discusses kernel regression and model selection to determine photometric redshifts astro-ph/0706.2704. This paper presents their studies on choosing bandwidth of kernels via 10 fold cross-validation, choosing appropriate models from various combination of input parameters through estimating root mean square error and AIC, and evaluating their kernel regression to other regression and classification methods with root mean square errors from literature survey. They made a conclusion of flexibility in kernel regression particularly for data at high z.
Continue reading ‘[ArXiv] Kernel Regression, June 20, 2007’ »