FWIW, HID is time invariant within state...
-
Login or Register
- Log in with
- You are not logged in. You can browse but not post. Login or Register by clicking 'Login or Register' at the top-right of this page. For more information on Statalist, see the FAQ.
-
-
Among the 18 state with HID == 0 do you have (in the estimation sample) both some with cutoff3 == 0 and some with cutoff3 == 1 in both the pre- and post periods?Comment
-
Assuming I understand what you're asking, of 18 states with HID==0, the same 18 have CUTOFF3==0 in both the pre- and post-periods; that doesn't change. Similarly, the same 8 states with HID==0 have CUTOFF3 status of 1. I'm trying to understand the influence of a state-level policy once the *federal* benefits have been turned off; so did the state-level policy wield greater influence once the federal benefits were turned off. So CUTOFF3==1 refers to treatment states -- those states that cut off benefits. Here's a breakdown of the states:
HTML Code:. tab statefip if hid==1 & cutoff3==0 & [postperiod==0] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- hawaii | 7,265 11.39 11.39 kansas | 6,566 10.29 21.68 michigan | 11,696 18.33 40.02 new mexico | 8,518 13.35 53.37 north carolina | 12,371 19.39 72.76 washington | 9,947 15.59 88.35 wisconsin | 7,429 11.65 100.00 ----------------------------------------+----------------------------------- Total | 63,792 100.00 . tab statefip if hid==1 & cutoff3==0 & [postperiod==1] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- hawaii | 2,433 11.51 11.51 kansas | 2,062 9.76 21.27 michigan | 3,979 18.83 40.10 new mexico | 2,895 13.70 53.79 north carolina | 4,129 19.54 73.33 washington | 3,230 15.28 88.62 wisconsin | 2,406 11.38 100.00 ----------------------------------------+----------------------------------- Total | 21,134 100.00 . tab statefip if hid==0 & cutoff3==0 & [postperiod==0] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- california | 43,750 23.45 23.45 colorado | 6,534 3.50 26.95 connecticut | 5,061 2.71 29.67 delaware | 5,955 3.19 32.86 district of columbia | 7,949 4.26 37.12 illinois | 13,751 7.37 44.49 kentucky | 6,270 3.36 47.85 maine | 4,168 2.23 50.08 maryland | 6,601 3.54 53.62 minnesota | 6,686 3.58 57.21 nevada | 7,068 3.79 61.00 new jersey | 9,540 5.11 66.11 new york | 18,952 10.16 76.27 oregon | 9,030 4.84 81.11 pennsylvania | 13,996 7.50 88.61 rhode island | 4,497 2.41 91.02 vermont | 7,264 3.89 94.91 virginia | 9,488 5.09 100.00 ----------------------------------------+----------------------------------- Total | 186,560 100.00 . tab statefip if hid==0 & cutoff3==0 & [postperiod==1] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- california | 14,189 23.61 23.61 colorado | 2,139 3.56 27.17 connecticut | 1,730 2.88 30.05 delaware | 1,854 3.09 33.14 district of columbia | 2,497 4.16 37.29 illinois | 4,575 7.61 44.90 kentucky | 2,113 3.52 48.42 maine | 1,394 2.32 50.74 maryland | 2,060 3.43 54.17 minnesota | 2,086 3.47 57.64 nevada | 2,233 3.72 61.36 new jersey | 3,135 5.22 66.57 new york | 6,073 10.11 76.68 oregon | 2,664 4.43 81.11 pennsylvania | 4,603 7.66 88.77 rhode island | 1,478 2.46 91.23 vermont | 2,335 3.89 95.12 virginia | 2,934 4.88 100.00 ----------------------------------------+----------------------------------- Total | 60,092 100.00 . tab statefip if hid==1 & cutoff3==1 & [postperiod==0] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- idaho | 9,335 10.36 10.36 mississippi | 9,372 10.40 20.75 missouri | 7,614 8.45 29.20 nebraska | 6,535 7.25 36.45 new hampshire | 7,027 7.80 44.25 north dakota | 7,994 8.87 53.12 oklahoma | 7,858 8.72 61.84 south carolina | 8,988 9.97 71.81 tennessee | 9,856 10.93 82.74 utah | 8,246 9.15 91.89 wyoming | 7,309 8.11 100.00 ----------------------------------------+----------------------------------- Total | 90,134 100.00 . tab statefip if hid==1 & cutoff3==1 & [postperiod==1] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- idaho | 2,896 9.93 9.93 mississippi | 2,915 10.00 19.93 missouri | 2,560 8.78 28.71 nebraska | 2,178 7.47 36.18 new hampshire | 2,343 8.04 44.22 north dakota | 2,688 9.22 53.44 oklahoma | 2,402 8.24 61.68 south carolina | 2,898 9.94 71.62 tennessee | 3,210 11.01 82.63 utah | 2,699 9.26 91.88 wyoming | 2,366 8.12 100.00 ----------------------------------------+----------------------------------- Total | 29,155 100.00 . tab statefip if hid==0 & cutoff3==1 & [postperiod==0] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- alabama | 11,297 12.38 12.38 arkansas | 9,438 10.34 22.72 georgia | 11,394 12.48 35.20 iowa | 5,711 6.26 41.46 montana | 9,284 10.17 51.63 south dakota | 5,818 6.37 58.01 texas | 28,047 30.73 88.73 west virginia | 10,282 11.27 100.00 ----------------------------------------+----------------------------------- Total | 91,271 100.00 . tab statefip if hid==0 & cutoff3==1 & [postperiod==1] state (fips code) | Freq. Percent Cum. ----------------------------------------+----------------------------------- alabama | 3,724 12.42 12.42 arkansas | 2,992 9.98 22.40 georgia | 3,689 12.30 34.71 iowa | 1,895 6.32 41.03 montana | 3,141 10.48 51.50 south dakota | 1,792 5.98 57.48 texas | 9,366 31.24 88.72 west virginia | 3,381 11.28 100.00 ----------------------------------------+----------------------------------- Total | 29,980 100.00Comment
-
Oh, HID is time invariant within state! Then, yes, some of these things are going to be omitted because the full 8 combinations of HID 0/1, CUTOFF3 0/1 and postperiod 0/1 are not all instantiated. So HID is not really a treatment at all, it's another state level attribute (and, therefore, also colinear with state). The bottom line here is that no matter how you slice it, you are not going to be able to disentangle the HID effect from the state effects in the same model. Getting rid of the state effects altogether (and all other variables time invariant within state except HID) would be the (typographically) simplest solution. Keeping the state effects in the model precludes estimating the effect of HID. If you feel that it is crucial to include some geographical effects I can think of two ways to do that, but they involve compromises.- You could go to a random effects logistic model with state as the random intercept. First you would -xtset state-, and then you would use -xtlogit reemp3 i.hid##i.cutoff3##i.postperiod ..., re vce(statefips)- where ... includes all of the covariates from your original -logit- model except statefips and any other variable that is time-invariant within state. The colinearity does not become an issue in random effects models. Now, caveat: many economists reject random effects models (sometimes for good reasons, but sometimes not). You may face difficulties getting this published in an economic journal if you go this route.
- You could stay with fixed effects, but instead of having a state-level fixed effect, create a new variable that indicates groups of states. One choice for this might be census region. Probably better from a substantive point of view is if you can find a variable that groups the states in ways that are relevant to employment, like based on decile of unemployment in 2018, or something like that. As you are working in this area, I'm sure you can come up with better ideas for an employment-relevant way to group states than I can. This will be not quite the same as a state effect, but as long as each of the state-groups contains some examples of all four combinations of hid and cutoff3, you will avoid colinearity issues and will have a separate estimate of the treatment effects.
Comment
-
Thank you very much Clyde. Okay, so *for now*, I might just remove state and month fixed effects from my models (or can I keep month FE?), because it's the effects of HID (and other state-level policy variables I'm working with in other analyses) that I'm most interested in. I have a series of state labor market slack measures, plus a host of individual-level controls. That might have to be sufficient for now. Then, when I have more time, I might try to implement this random effects model. :/ :/Comment
-
Well, the time variables are problematic because they are colinear with the postperiod variable. You can't have both in the same model.Comment
-
Right; okay. Thank you so much again. I'm sure I'll have more questions as I continue this work!Comment
-
Clyde, I posted a question in a new thread on the margins command, with an interaction of a factor and continuous logged variable. It's not unrelated to the discussion here. Any thoughts? Thanks.
https://www.statalist.org/forums/for...ogged-variableComment
Comment