User:EpochFail/Journal/2011-12-12
< User:EpochFail | Journal
Tuesday, Dec. 13th[edit]
I'm updating my tests from last week to simple Chi^2 comparisons with the control.
Below, each "diff" represents the proportion difference in the outcomes between the experimental case and the control along with a "p-value" and "conf.int" (95% two tailed confidence interval) for the proportion difference test.
Result's for 420 'vandal' editors: ============================================================ Good outcome: Control : prop= 0.617, n=128 Personal : prop= 0.590, diff=-0.027, p-value= 0.742, conf.int=(-0.152, 0.098), n=139 No Directives: prop= 0.550, diff=-0.068, p-value= 0.309, conf.int=(-0.190, 0.055), n=151 Improves: Control : prop= 0.529, n=70 Personal : prop= 0.468, diff=-0.061, p-value= 0.566, conf.int=(-0.236, 0.114), n=77 No Directives: prop= 0.489, diff=-0.040, p-value= 0.735, conf.int=(-0.208, 0.129), n=90 Contact: Control : prop= 0.039, n=128 Personal : prop= 0.058, diff= 0.018, p-value= 0.677, conf.int=(-0.040, 0.077), n=139 No Directives: prop= 0.053, diff= 0.014, p-value= 0.791, conf.int=(-0.042, 0.070), n=151 Stays: Control : prop= 0.547, n=128 Personal : prop= 0.554, diff= 0.007, p-value= 1.000, conf.int=(-0.119, 0.134), n=139 No Directives: prop= 0.596, diff= 0.049, p-value= 0.480, conf.int=(-0.075, 0.173), n=151 Good contact: Control : prop= 0.400, n=5 Personal : prop= 0.625, diff= 0.225, p-value= 0.826, conf.int=(-0.482, 0.932), n=8 No Directives: prop= 0.750, diff= 0.350, p-value= 0.499, conf.int=(-0.336, 1.000), n=8 Result's for 982 'bad faith' editors: ============================================================ Good outcome: Control : prop= 0.697, n=343 Personal : prop= 0.728, diff= 0.031, p-value= 0.421, conf.int=(-0.041, 0.103), n=320 No Directives: prop= 0.688, diff=-0.009, p-value= 0.866, conf.int=(-0.083, 0.064), n=317 Improves: Control : prop= 0.272, n=147 Personal : prop= 0.353, diff= 0.080, p-value= 0.180, conf.int=(-0.034, 0.195), n=139 No Directives: prop= 0.293, diff= 0.020, p-value= 0.796, conf.int=(-0.089, 0.130), n=147 Contact: Control : prop= 0.017, n=343 Personal : prop= 0.038, diff= 0.020, p-value= 0.179, conf.int=(-0.008, 0.048), n=320 No Directives: prop= 0.041, diff= 0.024, p-value= 0.116, conf.int=(-0.005, 0.052), n=317 Stays: Control : prop= 0.429, n=343 Personal : prop= 0.434, diff= 0.006, p-value= 0.942, conf.int=(-0.073, 0.084), n=320 No Directives: prop= 0.464, diff= 0.035, p-value= 0.407, conf.int=(-0.044, 0.114), n=317 Good contact: Control : prop= 0.667, n=6 Personal : prop= 0.500, diff=-0.167, p-value= 0.867, conf.int=(-0.763, 0.430), n=12 No Directives: prop= 0.538, diff=-0.128, p-value= 0.979, conf.int=(-0.714, 0.458), n=13 Result's for 702 'test' editors: ============================================================ Good outcome: Control : prop= 0.142, n=247 Personal : prop= 0.072, diff=-0.070, p-value= 0.020, conf.int=(-0.128, -0.011), n=236 No Directives: prop= 0.111, diff=-0.031, p-value= 0.388, conf.int=(-0.096, 0.034), n=217 Improves: Control : prop= 0.340, n=100 Personal : prop= 0.250, diff=-0.090, p-value= 0.283, conf.int=(-0.241, 0.061), n=68 No Directives: prop= 0.196, diff=-0.144, p-value= 0.037, conf.int=(-0.278, -0.011), n=92 Contact: Control : prop= 0.016, n=247 Personal : prop= 0.021, diff= 0.005, p-value= 0.945, conf.int=(-0.023, 0.033), n=236 No Directives: prop= 0.074, diff= 0.058, p-value= 0.005, conf.int=( 0.015, 0.100), n=217 Stays: Control : prop= 0.405, n=247 Personal : prop= 0.288, diff=-0.117, p-value= 0.009, conf.int=(-0.205, -0.028), n=236 No Directives: prop= 0.424, diff= 0.019, p-value= 0.747, conf.int=(-0.075, 0.113), n=217 Good contact: Control : prop= 0.500, n=4 Personal : prop= 1.000, diff= 0.500, p-value= 0.324, conf.int=(-0.215, 1.000), n=5 No Directives: prop= 0.688, diff= 0.188, p-value= 0.907, conf.int=(-0.509, 0.884), n=16 Result's for 347 'good faith' editors: ============================================================ Good outcome: Control : prop= 0.388, n=116 Personal : prop= 0.387, diff=-0.001, p-value= 1.000, conf.int=(-0.128, 0.127), n=111 No Directives: prop= 0.398, diff= 0.010, p-value= 0.977, conf.int=(-0.123, 0.144), n=118 Improves: Control : prop= 0.230, n=61 Personal : prop= 0.111, diff=-0.118, p-value= 0.129, conf.int=(-0.266, 0.029), n=63 No Directives: prop= 0.231, diff= 0.001, p-value= 1.000, conf.int=(-0.147, 0.150), n=65 Contact: Control : prop= 0.078, n=116 Personal : prop= 0.072, diff=-0.006, p-value= 1.000, conf.int=(-0.079, 0.068), n=111 No Directives: prop= 0.153, diff= 0.075, p-value= 0.112, conf.int=(-0.015, 0.165), n=118 Stays: Control : prop= 0.526, n=116 Personal : prop= 0.568, diff= 0.042, p-value= 0.619, conf.int=(-0.097, 0.180), n=111 No Directives: prop= 0.551, diff= 0.025, p-value= 0.801, conf.int=(-0.111, 0.161), n=118 Good contact: Control : prop= 1.000, n=9 Personal : prop= 0.750, diff=-0.250, p-value= 0.399, conf.int=(-0.668, 0.168), n=8 No Directives: prop= 0.944, diff=-0.056, p-value= 1.000, conf.int=(-0.217, 0.106), n=18
And now the same analysis of the old huggling dataset.
Result's for 174 'unlikely' editors: ============================================================ Good outcome: Control : prop= 0.533, n=45 Personal : prop= 0.688, diff= 0.154, p-value= 0.189, conf.int=(-0.063, 0.372), n=48 Teaching : prop= 0.771, diff= 0.238, p-value= 0.050, conf.int=( 0.011, 0.465), n=35 Personal & Teaching: prop= 0.771, diff= 0.238, p-value= 0.050, conf.int=( 0.011, 0.465), n=35 Improves: Control : prop= 0.208, n=24 Personal : prop= 0.238, diff= 0.030, p-value= 1.000, conf.int=(-0.244, 0.304), n=21 Teaching : prop= 0.300, diff= 0.092, p-value= 0.896, conf.int=(-0.306, 0.490), n=10 Personal & Teaching: prop= 0.300, diff= 0.092, p-value= 0.896, conf.int=(-0.306, 0.490), n=10 Contact: Control : prop= 0.067, n=45 Personal : prop= 0.104, diff= 0.038, p-value= 0.784, conf.int=(-0.097, 0.172), n=48 Teaching : prop= 0.000, diff=-0.067, p-value= 0.335, conf.int=(-0.165, 0.032), n=35 Personal & Teaching: prop= 0.000, diff=-0.067, p-value= 0.335, conf.int=(-0.165, 0.032), n=35 Stays: Control : prop= 0.533, n=45 Personal : prop= 0.438, diff=-0.096, p-value= 0.474, conf.int=(-0.320, 0.128), n=48 Teaching : prop= 0.286, diff=-0.248, p-value= 0.046, conf.int=(-0.482, -0.013), n=35 Personal & Teaching: prop= 0.286, diff=-0.248, p-value= 0.046, conf.int=(-0.482, -0.013), n=35 Good contact: Control : prop= 0.333, n=3 Personal : prop= 1.000, diff= 0.667, p-value= 0.206, conf.int=(-0.133, 1.000), n=5 Teaching : prop= --- , diff= --- , p-value= --- , conf.int=( --- , --- ), n=0 Personal & Teaching: prop= --- , diff= --- , p-value= --- , conf.int=( --- , --- ), n=0 Result's for 875 'possible' editors: ============================================================ Good outcome: Control : prop= 0.465, n=200 Personal : prop= 0.465, diff= 0.000, p-value= 1.000, conf.int=(-0.096, 0.096), n=215 Teaching : prop= 0.449, diff=-0.016, p-value= 0.803, conf.int=(-0.114, 0.081), n=243 Personal & Teaching: prop= 0.449, diff=-0.016, p-value= 0.803, conf.int=(-0.114, 0.081), n=243 Improves: Control : prop= 0.214, n=56 Personal : prop= 0.218, diff= 0.004, p-value= 1.000, conf.int=(-0.141, 0.149), n=78 Teaching : prop= 0.233, diff= 0.018, p-value= 0.961, conf.int=(-0.136, 0.173), n=86 Personal & Teaching: prop= 0.233, diff= 0.018, p-value= 0.961, conf.int=(-0.136, 0.173), n=86 Contact: Control : prop= 0.030, n=200 Personal : prop= 0.079, diff= 0.049, p-value= 0.049, conf.int=( 0.001, 0.097), n=215 Teaching : prop= 0.041, diff= 0.011, p-value= 0.711, conf.int=(-0.028, 0.050), n=243 Personal & Teaching: prop= 0.041, diff= 0.011, p-value= 0.711, conf.int=(-0.028, 0.050), n=243 Stays: Control : prop= 0.280, n=200 Personal : prop= 0.363, diff= 0.083, p-value= 0.090, conf.int=(-0.011, 0.177), n=215 Teaching : prop= 0.354, diff= 0.074, p-value= 0.120, conf.int=(-0.017, 0.165), n=243 Personal & Teaching: prop= 0.354, diff= 0.074, p-value= 0.120, conf.int=(-0.017, 0.165), n=243 Good contact: Control : prop= 0.667, n=6 Personal : prop= 0.588, diff=-0.078, p-value= 1.000, conf.int=(-0.601, 0.444), n=17 Teaching : prop= 0.600, diff=-0.067, p-value= 1.000, conf.int=(-0.618, 0.484), n=10 Personal & Teaching: prop= 0.600, diff=-0.067, p-value= 1.000, conf.int=(-0.618, 0.484), n=10 Result's for 116 'golden' editors: ============================================================ Good outcome: Control : prop= 0.500, n=32 Personal : prop= 0.588, diff= 0.088, p-value= 0.637, conf.int=(-0.182, 0.358), n=34 Teaching : prop= 0.391, diff=-0.109, p-value= 0.600, conf.int=(-0.410, 0.193), n=23 Personal & Teaching: prop= 0.391, diff=-0.109, p-value= 0.600, conf.int=(-0.410, 0.193), n=23 Improves: Control : prop= 0.050, n=20 Personal : prop= 0.048, diff=-0.002, p-value= 1.000, conf.int=(-0.137, 0.132), n=21 Teaching : prop= 0.000, diff=-0.050, p-value= 1.000, conf.int=(-0.196, 0.096), n=10 Personal & Teaching: prop= 0.000, diff=-0.050, p-value= 1.000, conf.int=(-0.196, 0.096), n=10 Contact: Control : prop= 0.125, n=32 Personal : prop= 0.206, diff= 0.081, p-value= 0.582, conf.int=(-0.127, 0.289), n=34 Teaching : prop= 0.000, diff=-0.125, p-value= 0.217, conf.int=(-0.277, 0.027), n=23 Personal & Teaching: prop= 0.000, diff=-0.125, p-value= 0.217, conf.int=(-0.277, 0.027), n=23 Stays: Control : prop= 0.625, n=32 Personal : prop= 0.618, diff=-0.007, p-value= 1.000, conf.int=(-0.249, 0.234), n=34 Teaching : prop= 0.435, diff=-0.190, p-value= 0.261, conf.int=(-0.491, 0.110), n=23 Personal & Teaching: prop= 0.435, diff=-0.190, p-value= 0.261, conf.int=(-0.491, 0.110), n=23 Good contact: Control : prop= 0.750, n=4 Personal : prop= 0.857, diff= 0.107, p-value= 1.000, conf.int=(-0.497, 0.712), n=7 Teaching : prop= --- , diff= --- , p-value= --- , conf.int=( --- , --- ), n=0 Personal & Teaching: prop= --- , diff= --- , p-value= --- , conf.int=( --- , --- ), n=0
Saturday, Dec. 17th[edit]
I want to be able to support Steven and Maryana and the experiments that they are doing with message templates. Recently, they requested that I help them put together some datasets. What follows are the fields they asked about:
- their edit count to all namespaces before and after warning
- overall edit count pre/post warning
- block/ban record pre/post warning
- number of edits to their talk page pre/post warning
- number of other warnings pre/post warning
- articles created pre/post warning
- deleted edits pre/post warning
- deleted articles pre/post warning
It looks like I be starting with a list of editors, though I may have to discover that too.
If I have a pattern I can match on the comment of the edit that leaves the message (e.g. "([[WP:HG|HG]])") as well as some content that can be matched within the message itself (e.g. "<!-- exp-template-foo -->" or "{{z78}}") I can find the message postings.
Figuring out when the message was read is a more difficult matter that we only have a half solution to at the moment. I'll ignore that for the time being since changing the pivoting moment from the time the message was posted to the time that the message was read should be trivial once we've solved the larger issue.
Now, there are a few different types of things that are being asked for, but they do all pivot around a user, at a timestamp.
I'm not sure how I want Steven and Maryana to interface with a system that gathers this data, but that's OK for now. No matter if it is a script they'll have to run or a web interface they can pull up, it will take the same parameters.
I think I'll start with exploring this problem with a python script.
17:34, 17 December 2011 (UTC)
I've got to call it quits for the day, but I have a pretty good idea of how to do this work somewhat efficiently with a combination of the slave database and the Wikipedia API (for getting texts). Time to ping Ryan.
18:10, 17 December 2011 (UTC)