{"items": [{"author": "David Chudzicki", "source_link": "https://plus.google.com/106120852580068301475", "anchor": "gp-1408512980240", "service": "gp", "text": "The analysis doesn't seem very helpful, but I disagree that PCA would help. The components you find may be interpretable, but not really more so than the regression as it stands. And if it's overfitting you're worried about, you should use regularization, which PCA is a not-so-good form of. \nhttp://en.m.wikipedia.org/wiki/Regularization_%28mathematics%29\n", "timestamp": 1408512980}, {"author": "David Chudzicki", "source_link": "https://plus.google.com/106120852580068301475", "anchor": "gp-1408513113566", "service": "gp", "text": "The principal components may be mainly the intuitive ones you expect, but you'll have to inspect the weights and think about if they really are. Which isn't any easier than carefully inspecting the coefficients in the original regression. ", "timestamp": 1408513113}, {"author": "Jeff Kaufman", "source_link": "https://plus.google.com/103013777355236494008", "anchor": "gp-1408630132381", "service": "gp", "text": "@David Chudzicki\n\u00a0The reason I'd like to see the author do PCA and throw away the low weighted components before continuing is that in the rest of their analysis they treat each variable as equally informative. \u00a0So if your model has ten slightly different ways of specifying X and one way of specifying Y, and you weight each way the same, you're going to end up paying too much attention to X and not enough to Y.", "timestamp": 1408630132}, {"author": "David Chudzicki", "source_link": "https://plus.google.com/106120852580068301475", "anchor": "gp-1408671545575", "service": "gp", "text": "Oh, jeez. I hadn't noticed that they are correlations rather than coefficients in a regression. Your criticism is spot on given what they are. I still doubt PCA would be my first choice among the many other approaches that could be used. ", "timestamp": 1408671545}, {"author": "David Chudzicki", "source_link": "https://plus.google.com/106120852580068301475", "anchor": "gp-1408671644881", "service": "gp", "text": "Given this approach (correlations rather than regression coefficients), yes, PCA would be a big help, since the components would be independent, so then it's not so\u00a0bad. (sorry I read too quickly)", "timestamp": 1408671644}, {"author": "BDan", "source_link": "https://plus.google.com/103775592027106438640", "anchor": "gp-1409002364144", "service": "gp", "text": "Having spent some time in smaller towns not in the Midwest, I think it's unlikely that they would be particularly walkable, unless the Midwest is unusual somehow. Small towns tend to cluster their commercial elements in a very small area, with residences spread out much more, and an overall low population density, such that most people will not actually be living that close to the center. In the tiniest towns, nearly everyone might live close to the center, but they'll also probably often lack many services, requiring a drive to a larger town to get them.", "timestamp": 1409002364}, {"author": "Jeff Kaufman", "source_link": "https://plus.google.com/103013777355236494008", "anchor": "gp-1409060544328", "service": "gp", "text": "@BDan\n\u00a0I think \"small town in the midwest\" is probably wrong but \"small city in the midwest\" might be right. \u00a0For example, I stayed with a couple in\u00a0indianapolis (near\u00a0\nhttp://goo.gl/maps/W9YCw\n) in a neighborhood that seemed very walkable (walkscrore gives them a 75) and much cheaper than boston (trulia gives $245k as the median listing price).", "timestamp": 1409060544}, {"author": "BDan", "source_link": "https://plus.google.com/103775592027106438640", "anchor": "gp-1409061487531", "service": "gp", "text": "@Jeff Kaufman\n Indianapolis isn't a small city, except maybe by comparison to very large cities. It's the 12th most populous city in the country, with a greater population than Boston, though much less densely packed, and with a smaller metropolitan area.", "timestamp": 1409061487}, {"author": "Jeff Kaufman", "source_link": "https://plus.google.com/103013777355236494008", "anchor": "gp-1409061851636", "service": "gp", "text": "@BDan\n\u00a0\"a greater population than Boston, though much less densely packed, and with a smaller metropolitan area.\"\n
\n
\nNominal city size is nearly useless, and mostly just tells you something about administration. \u00a0But indianapolis is #33 on\u00a0\nhttp://en.wikipedia.org/wiki/List_of_Metropolitan_Statistical_Areas\n at 2M people. \u00a0So yes, not small. \u00a0Maybe I just misinterpreted the author here?", "timestamp": 1409061851}]}