Research talk:Reading time/Work log/2018-11-09
Friday, November 9, 2018[edit]
I fit M2 and M3.
Regression Table[edit]
| model 2 | model 3 | ||
|---|---|---|---|
| Intercept | 10.3827 (0.0092)*** | 10.7961 (0.0108)*** | |
| mobile | -0.2019 (0.0011)*** | -0.9194 (0.0099)*** | |
| Human Development Index | -0.5861 (0.0058)*** | -1.0662 (0.0088)*** | |
| mobile : HDI | 0.8480 (0.0117)*** | ||
| Revision length (bytes) | 0.1665 (0.0005)*** | 0.1665 (0.0005)*** | |
| time to first paint | -0.0000 (0.0000)*** | -0.0000 (0.0000)*** | |
| time to dom interactive | 0.0000 (0.0000)*** | 0.0000 (0.0000)*** | |
| sessionlength | -0.0001 (0.0000)*** | -0.0001 (0.0000)*** | |
| lastinsessionTRUE | 0.6180 (0.0011)*** | 0.6159 (0.0011)*** | |
| nthinsession | 0.0002 (0.0000)*** | 0.0002 (0.0000)*** | |
| dayofweekMon | 0.0959 (0.0020)*** | 0.0959 (0.0020)*** | |
| dayofweekSat | 0.0087 (0.0020)*** | 0.0064 (0.0020)** | |
| dayofweekSun | 0.0269 (0.0020)*** | 0.0253 (0.0020)*** | |
| dayofweekThu | 0.0489 (0.0020)*** | 0.0489 (0.0020)*** | |
| dayofweekTue | 0.0291 (0.0020)*** | 0.0294 (0.0020)*** | |
| dayofweekWed | 0.0679 (0.0020)*** | 0.0677 (0.0020)*** | |
| usermonth4 | 0.0102 (0.0097) | 0.0104 (0.0097) | |
| usermonth5 | 0.0127 (0.0096) | 0.0124 (0.0096) | |
| usermonth6 | -0.0072 (0.0099) | -0.0076 (0.0098) | |
| usermonth7 | -0.0465 (0.0098)*** | -0.0474 (0.0098)*** | |
| usermonth8 | -0.0103 (0.0098) | -0.0111 (0.0098) | |
| usermonth9 | 0.0425 (0.0077)*** | 0.0432 (0.0077)*** | |
| usermonth10 | -0.0034 (0.0076) | -0.0035 (0.0076) | |
| R2 | 0.0520 | 0.0525 | |
| Adj. R2 | 0.0520 | 0.0525 | |
| Num. obs. | 9873641 | 9873641 | |
| RMSE | 14.3858 | 14.3819 | |
| ***p < 0.001, **p < 0.01, *p < 0.05 | |||
Plot for interpreting the interaction term[edit]
Density of English Wikipedia Page Visible Times (Logged). This chart shows marginal effects plot for the interaction between mobile phone usage and the human development index (HDI) of the country the reader is in. It shows that prototypical readers on the mobile site read for shorter times and that prototypical readers in lower HDI countries read for longer times, and that this gap is mostly on non-mobile devices.
Model 3 shows an interesting result: readers in lower HDI contexts are likely to read for longer than readers in higher HDI contexts, but the difference is virtually only on the desktop site not the mobile site. To me this looks like evidence in support of the "device gap" theory, which says that inherent limitations of phones (i.e. small and low quality screens, slow processors, battery life) make them inferior for more complex tasks. Of course there are alternative explanations such as the situations people might find themselves where they have access to a phone but not a desktop (e.g. looking up trivia in a conversation).