Timeline for Hypothesis test on data with confounding spatial clustering
Current License: CC BY-SA 3.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 21, 2013 at 22:22 | history | bounty ended | ali_m | ||
| Apr 21, 2013 at 22:21 | vote | accept | ali_m | ||
| Apr 19, 2013 at 19:25 | comment | added | ali_m | Yeah, so to complicate things a bit further I have 2-3 planes taken ad different depths from a 3D volume of brain tissue, and the planes are unevenly and sparsely sampled in Z. Some pairs of my 'treatment' cells fall within the same plane but most are across planes. | |
| Apr 19, 2013 at 18:55 | comment | added | generic_user | Are you in 3d space then? Or only looking at the surface of the brain? I'm sure that you could probably generalize the "Coneley estimator" to 3d space, but I haven't seen it done. If you can pull it off you should publish it somewhere. Note that the code takes geographic coordinates as input, which are modified into distance based on trigonometry. | |
| Apr 19, 2013 at 10:40 | comment | added | ali_m | Sol uses a uniform kernel. I really don't have a good idea of what sorts of spatial shocks I should expect and the scales over which they might operate (to give you a bit of background I'm looking at the responses of neurons within a network where I expect nearby neurons to receive more similar inputs, but I know nothing about the fine-scale network structure). As you suggest, I might try a range of different bandwidths and see what they do to the standard errors. | |
| Apr 18, 2013 at 18:34 | comment | added | generic_user | That is true. I forget whether Sol implements a gaussian or uniform or what kind of kernel. Are you truly in a situation where you are completely uninformed about what sorts of unobserved spatial shocks are possible and what scales they might operate over? Could you specify several different bandwidths, and maybe plot the standard errors as a function of bandwidth to see how your result is sensitive to the choice, then perhaps use the most conservative bandwidth that is logically consistent with what the data generating process might be? | |
| Apr 18, 2013 at 14:58 | comment | added | ali_m | But based on Sol's code I would still have to select a spatial kernel with some specified cutoff or bandwidth. When I choose the bandwidth of the kernel I'm making an implicit assumption about the spatial scale over which I expect my response values to be correlated. | |
| Apr 18, 2013 at 3:18 | comment | added | generic_user | Space doesn't come into it until you try to estimate the variance of your estimator. Intuitively, if all the difference is close together, you're less certain that your estimate isn't just a relic of some unobserved spatial shock. | |
| Apr 18, 2013 at 3:17 | comment | added | generic_user | Theoretical econometricians unfortunately seem to take pleasure in obfuscating. Basically what it says is run whatever regression you want, and then go fix the standard errors later (i.e.: using Sol's code). You say you don't want to specify a linear model. Why not specify a really simple one? Y= a + bT + e? where T is your treatment? This is the same thing as a T test, with the added feature that you could tack stuff onto it if you change your mind. | |
| Apr 18, 2013 at 1:18 | comment | added | ali_m | Thanks for response, although I must admit that I found the references rather heavy going as a non-statistician. My (vague!) understanding is that what you're suggesting would require me to form a linear regression model relating differences in treatment and spatial distance to differences in response. I'd prefer to avoid forming explicit linear models if possible, since it's very likely that the relationship between distance and response similarity is roughly linear only over a restricted range of distances. Can you think of a way to test this hypothesis without resorting to linear models? | |
| Apr 15, 2013 at 1:32 | history | answered | generic_user | CC BY-SA 3.0 |