Dear Statalist,

I am estimating the effect of parental separation on the amount of time a child spends with both biological parents simultaneously. We would expect when parents separate, children drop "two-parent time" to nearly zero. And that is what I find. However, I also looked at how "two-parent time" evolves during the years after parental separation, tracing up to 6 years later. We could expect a slight increase because normally biological parents start seeing each other a bit more once the breakup's stress/grief/etc has passed or calmed down. But I find that after "two-parent time" dropping to nearly 0 minutes a day in the first observation after separation, then it increases to an amount of time that it is difficult to believe. So then I controlled for a variable capturing whether the mother (who typically stays with the child after separation) starts a new relationship with another man, and that should explain the increase in "two-parent time". Indeed the descriptives show that two-parent time is way more when the mother finds a new partner, compared to when the mother is single, which makes total sense. We are talking about 7-10 minutes on average per day of two-parent time when the mother is single, compared to 65-130 minutes a day on average when the mother re-partners (the range represents the year of observation after separation). Here the analyses:

Code:
xtreg both i.timex ib1.rep_age moth_degree father_degree siblings english i.state rem  if day==1, fe
margins timex, post noestimcheck
marginsplot, title("Two-Parents Time") ylabel(0(100)300) ///
yscale(range(-15 (75) 375)) ///
graphregion(color(white)) ///
plot1opts(msymbol(s) mcolor(gr8)lwidth(thin)) ///
ciopts(recast(rcap) lcolor(gr8) lwidth(thin))
graph save a3 , replace
(both = two-parent time; timex=time since separation; rep_age=child's age; ... rem=mother re-partnered)

I get these results:

Code:
note: 1b.rep_age identifies no observations in the sample
note: 14.rep_age omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =     14,865
Group variable: hicid                           Number of groups  =      3,721

R-sq:                                           Obs per group:
     within  = 0.1790                                         min =          1
     between = 0.0567                                         avg =        4.0
     overall = 0.1241                                         max =          6

                                                F(22,11122)       =     110.22
corr(u_i, Xb)  = -0.0947                        Prob > F          =     0.0000

-------------------------------------------------------------------------------
         both |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
        timex |
       -4/-6  |  -44.29691   28.51901    -1.55   0.120    -100.1992    11.60541
          -2  |  -39.79602   30.55646    -1.30   0.193    -99.69209    20.10005
           0  |  -158.4857   30.66365    -5.17   0.000    -218.5919   -98.37952
          +2  |  -125.0832   32.80271    -3.81   0.000    -189.3824   -60.78411
       +4/+6  |  -92.93226   33.57639    -2.77   0.006    -158.7479   -27.11659
              |
      rep_age |
           1  |          0  (empty)
           4  |   167.4898   6.545673    25.59   0.000     154.6591    180.3205
           6  |    135.185   6.418732    21.06   0.000     122.6031    147.7668
           8  |   175.9095   6.501444    27.06   0.000     163.1656    188.6535
          10  |  -9.225281   6.295141    -1.47   0.143    -21.56487     3.11431
          12  |  -.1766192   6.343757    -0.03   0.978    -12.61151    12.25827
          14  |          0  (omitted)
              |
  moth_degree |   18.59104   14.86358     1.25   0.211     -10.5442    47.72629
father_degree |    -11.235   17.87196    -0.63   0.530     -46.2672     23.7972
     siblings |  -1.068718   9.012293    -0.12   0.906    -18.73441    16.59697
      english |  -19.51602   33.29769    -0.59   0.558     -84.7854    45.75336
              |
        state |
        VIC   |   62.18712   28.84275     2.16   0.031     5.650211     118.724
        QLD   |   41.34557   24.34281     1.70   0.089    -6.370648    89.06179
         SA   |   37.12378   36.57965     1.01   0.310    -34.57881    108.8264
         WA   |   51.92697   33.87054     1.53   0.125    -14.46528    118.3192
        TAS   |  -.5626034   49.98316    -0.01   0.991    -98.53847    97.41326
         NT   |   15.09654   45.20512     0.33   0.738    -73.51351    103.7066
        ACT   |    4.79004   35.14362     0.14   0.892    -64.09769    73.67777
              |
          rem |   65.72277   22.23282     2.96   0.003      22.1425     109.303
        _cons |   65.15493   35.34026     1.84   0.065    -4.118244    134.4281
--------------+----------------------------------------------------------------
      sigma_u |     149.78
      sigma_e |  199.42452
          rho |  .36065175   (fraction of variance due to u_i)
-------------------------------------------------------------------------------
F test that all u_i=0: F(3720, 11122) = 1.62                 Prob > F = 0.0000

. margins timex, post noestimcheck

Predictive margins                              Number of obs     =     14,865
Model VCE    : Conventional

Expression   : Linear prediction, predict()

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       timex |
     -8/-10  |   164.1324   4.190609    39.17   0.000     155.9189    172.3458
      -4/-6  |   119.8355   24.87412     4.82   0.000     71.08309    168.5879
         -2  |   124.3364   27.02868     4.60   0.000     71.36112    177.3116
          0  |   5.646679   27.07686     0.21   0.835      -47.423    58.71635
         +2  |   39.04915    29.3349     1.33   0.183     -18.4462    96.54449
      +4/+6  |   71.20013   30.13596     2.36   0.018     12.13474    130.2655
------------------------------------------------------------------------------
The result that the child spends around 80 minutes a day with two parents 6 years after separation is implausible. That should be explained because the mother found a new partner. And the descriptives indeed show that. But regression results are the same, with or without controlling for mother's new partnership - how is that possible and how can I solve this issue?

If you need more code, results or analyses just let me know and I happy to paste them here.

Thank you so so much!