Ask the grumpies: Time spent on housework by child status and gender

Laura Vanderkam asks:

Looking at the ATUS, how does having a kid affect how much time people devote to housework? Is this different for men and women? There are lots of different stories one could come up with: everyone does more housework because there’s more housework to be done. Everyone does less housework because there’s less available time to do it in. Mom does more and dad does less because they wind up conforming to traditional gender roles (and maybe mom winds up working less for pay, and so is the one around to do it). Maybe mom does a lot more and dad does a little more. So I’d love to know what the numbers actually show.

Lalalalala, Stata.  Ok, so I’m using the 2002-2012 ATUS here because I’m too lazy to download the 2013 one even though it’s now available.  In a bit I’ll show how things have changed if you limit to just 2011 and 2012.

How does having a kid affect how much time people devote to housework:

. ttest  bls_hhact_hwork, by(hh_child)

Two-sample t test with equal variances
——————————————————————————
Group |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
———+——————————————————————–
No |   72370     39.6372    .3054511    82.17145    39.03851    40.23588
Yes |   64590    44.33705    .3295443    83.75224    43.69114    44.98296
———+——————————————————————–
combined |  136960    41.85364    .2241497    82.95358    41.41431    42.29297
———+——————————————————————–
diff |           -4.699851    .4488466               -5.579582    -3.82012
——————————————————————————
diff = mean(No) – mean(Yes)                                   t = -10.4710
Ho: diff = 0                                     degrees of freedom =   136958

Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
Pr(T < t) = 0.0000         Pr(|T| > |t|) = 0.0000          Pr(T > t) = 1.0000

Urgh, I can’t figure out how to make this pretty without making it a picture and I’m too lazy to do that (in word you make it courier new 9 or smaller and it’s all pretty).  Anyhow, this is saying that people with kids spend 44.33 minutes on housework and people without kids spend 39.63 minutes on housework during the reference day.  This is a difference of 4.7 minutes.  This difference (two-tailed is the one in the middle, since we didn’t have a prior about which direction it should go) is significant at the 5% level (also at the .0001% level).  So kids create housework.  (Which is no surprise, but the surprise is that people spend time doing housework– childcare is measured under a different variable.)

Note that theologyandgeometry reminded me that I’m supposed to be using sampling weights when I do this, and they do matter somewhat in the regression results.  Unfortunately, ttest doesn’t take weights.   The kludge is a pain in the rear in Stata 11 (which is what I have on my home computer), so I apologize, but you’re getting the unweighted results.

Next:  Is this different for men and women?

Let’s say I want to answer this question in one fell swoop.  I would do a regression with an interaction.  It would look something like this:

unweighted:
Housework_min = 18.96 + 37.47*Female – 1.04*hh_child + 8.21*(Femalehh_child)

I can’t get the standard errors to line up in wordpress, but the se for the intercept is 0.31, se for Female is 0.57, se for hh_child is 0.44, se for the interaction term is 0.82.   To see whether these coefficients are significant, you take the coeff and divide by the standard error to get the p-value.  If that number is bigger than 1.96, it is significant at the 5% level.  These coefficients are all significant.

weighted to take into account sampling weights:
Housework_min = 15.47 + 38.50*Female – 0.67*hh_child + 4.06*(Femalehh_child)

Here everything is significant at the 1% level except the main effect on hh_child is no longer significant even at the 10% level, with a se of 0.49.  So weights do matter.  Thanks for reminding me, theologyandgeometry!

Ok, so what does this regression *mean*?  Plug and chug, my dear Watson, plug and chug.

The way the dataset is coded, if you’re female, Female is coded as 1.  If you’re not female, then it is coded as 0 (it doesn’t allow for female and not female at the same time).  Similarly, hh_child is one if you have a child under age 18 in the household and 0 if you don’t.

So to answer: “how does having a kid affect how much time people devote to housework?” You would take [18.96 + 37.47*Female – 1.04*hh_child + 8.21*(Femalehh_child)] and plug in 1 for hh_child and then plug in 0 for hh_child.

[18.96 + 37.47*Female – 1.04 + 8.21*(Female)] – [18.96 + 37.47*Female – 0 + 0)]

The 18.96 drops out, the 37.47 drops out, and you’re left with -1.04 + 8.21*Female.

For women:  [-1.04 + 8.21*1] => having kids correlates with 7.17 minutes more housework

For men:  [-1.04 + 0]  => having kids correlates with 1.04 minutes less of housework

The savvy econometrician will note here that we’ve seen these numbers before– that -1.04 is the coefficient for the hh_child variable, and the 7.17 is what you get if you add that coefficient to the interaction term.

Doing the weighted version, you get:

For women: [-0.67+4.06*1] = having kids correlates with 3.39 minutes more housework

For men:  [-0.67+0] => having kids correlates with 0.67 minutes less of housework

Now, one concern is that there are a lot more single parent households with women heads than with men.  Let’s see what happens when we limit to married households with both spouses present only.

ttest  bls_hhact_hwork if married==1, by(hh_child)

Two-sample t test with equal variances
——————————————————————————
Group |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
———+——————————————————————–
0. No |   27875    40.94371    .5036533    84.08898    39.95653     41.9309
1. Yes |   40403    46.88803    .4207222    84.56725     46.0634    47.71265
———+——————————————————————–
combined |   68278    44.46122    .3230849    84.42228    43.82797    45.09446
———+——————————————————————–
diff |           -5.944315    .6569407               -7.231918   -4.656712
——————————————————————————
diff = mean(0. No) – mean(1. Yes)                             t =  -9.0485
Ho: diff = 0                                     degrees of freedom =    68276

Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
Pr(T < t) = 0.0000         Pr(|T| > |t|) = 0.0000          Pr(T > t) = 1.0000

Having a child makes time spent on housework go up even more for two parent households than it does for everybody (about 6 minutes).  The difference is about one minute for unmarried households.  Maybe dads make a lot of mess.  More likely single moms don’t have time to do additional household chores while single people do have more time.  (Doing the interaction, this difference in the effect of having children between married and single couples is significant.)

Limiting to married couples only:

Housework_min = 14.23 + 51.44*Female +4.14*hh_child + 2.31*(Femalehh_child)

The interaction term is only marginally significant, and note a sign change on the hh_child coefficient.  Having a child affects married people by 4.14 +2.31*female.  Married men’s housework goes up by 4.14 minutes after having a child, but married women’s goes up by 6.45 minutes.

When you do it weighted, everything is significant at the 5% level.

Housework_min = 13.16+ 50.95*Female +2.51*hh_child + 3.39*(Femalehh_child)

Having a child affects married people by 2.51 + 3.39*female.  Married men’s housework goes up by 2.51 minutes after having a child, but a married woman’s goes up by 5.9 minutes.

Limiting to unmarried people only:

Housework_min = 22.29 + 28.89*Female – 5.41*hh_child + 6.34*(Femalehh_child)

All coefficients are significant.  Having a child affects unmarried people by -5.41 + 6.34*female.  Unmarried men’s housework goes down by 5.41 minutes and Unmarried women’s goes up by 6.34 minutes.  (Note that there are ~8,000 single men with kids and 16,000 single women with kids here, though I’m including married people whose spouses are absent in the “not married” category because we’re talking about housework.  It is more standard to include them in the married category when you’re looking at outcomes we care about like child well-being.)

Weighting the unmarried people regression:

Housework_min = 17.68 + 26.93*Female – 4.36*hh_child + 1.60*(Femalehh_child)

Here the interaction term is no longer significant, which suggests there isn’t a difference by gender in terms how how having a child affects housework.  Makes me wonder who the sampling frame is over- or under- sampling!  Here having a child affects unmarried people by -4.36 + 1.60*female.  Unmarried men’s housework goes down by 4.36 minutes when having a child and unmarried women’s also goes down (!) by 2.76 minutes.

There are other cuts that can be made… by age, by race, by ethnicity, by education, by work status etc.

I’m going to look now at the most recent years, 2011 and 2012.  Men are supposed to be more equal partners these days so…

. ttest  bls_hhact_hwork if year>2010, by(hh_child)

Two-sample t test with equal variances
——————————————————————————
Group |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
———+——————————————————————–
0. No |   13862    38.72681     .700486    82.47311    37.35376    40.09985
1. Yes |   11060    44.37197    .8187207    86.10202    42.76713    45.97681
———+——————————————————————–
combined |   24922    41.23204    .5330305    84.14795    40.18727    42.27682
———+——————————————————————–
diff |           -5.645164    1.072289               -7.746914   -3.543414
——————————————————————————
diff = mean(0. No) – mean(1. Yes)                             t =  -5.2646
Ho: diff = 0                                     degrees of freedom =    24920

Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
Pr(T < t) = 0.0000         Pr(|T| > |t|) = 0.0000          Pr(T > t) = 1.0000

Having a child still increases the amount of housework done by around 5.6 minutes (so more than for the 10-year period).

Housework_min = 20.34 + 33.75*Female – 0.42*hh_child + 8.68*(Femalehh_child)

Here the coefficient on hh_child is nowhere near significant.  The interaction term is still significant, but having a child has no significant effect on minutes worked by itself, only as it interacts with gender.  Men no longer work less when they have a child.  But women still work more!  Results are pretty similar with the weights.

Limiting to married only provides:

Housework_min = 16.35 + 46.28*Female + 4.04*hh_child + 2.50*(Femalehh_child)

Now hh_child is significant, but the interaction term is no longer significant!  Everyone in a married couple works 4 min more (you could argue that women work 6 min more, but that difference is not significant) once they have children.  Again the weights matter, because with them, you get:

Housework_min = 15.06 + 44.79*Female + 1.55*hh_child + 6.22*(Femalehh_child)

With the weights, hh_child is back to being no longer significant and the interaction term is significant at the 10% level.   Married women work marginally significantly more than married women do upon birth of a child.

Limiting to the unmarried (and those with absent spouses) provides:

Housework_min = 22.77 + 27.25*Female – 3.89*hh_child + 7.72*(Femalehh_child)

These are all significant.  Having a child decreases the amount of housework for unmarried men by 4 minutes, but increases it for unmarried women by around 4 minutes.  (These results hold if I drop people who are married with spouse absent, so it’s not like truckers are driving this result.)

Putting the weights in again changes things:

Housework_min =18.15 + 26.06*Female – 3.04*hh_child + 1.41*(Femalehh_child)

Female is significant (as is the constant) but the other terms are not.  This argues that there’s really no difference once you have a kid in how much housework you do if unmarried, either by gender or not.  It could be that there’s not enough unmarried fathers in the sample to say much of anything once the weights are added (perhaps they over-sample single dads, who knows!  Well, presumably ATUS knows.)  Also I should note that their sampling weights seem to be based on 2006 methodology, so if things have changed, they could be introducing measurement error which might tend to bias towards not finding anything.

All in all, there’s less significance with only the last two years of the data, but the story is still very similar.

So, to summarize:  Having kids increases the amount of housework that people do each day by 5-6 minutes on average, but about 1 minute for single-parent households.  On average, having kids means more housework for women and less housework for men.  However, in dual-parent married households with both spouses present, having a child increases rather than decreases the amount of time spent on housework for men.  In households with only one parent present, women do more housework and men do less (though with weighting it seems they both do less).  Potential reasons for this difference could be that men outsource the housework or that they’re more likely to substitute childcare for housework (or that they put their kids to work and women don’t!).

Now, the variable I used above assumes marriage.  It turns out there’s a variable in the ATUS that also gets at whether or not there’s an unmarried partner in the household.

tab spousepres

Spouse or unmarried partner in |
household |      Freq.     Percent        Cum.
—————————————-+———————————–
1. Spouse present |     69,359       50.64       50.64
2. Unmarried partner present |      4,224        3.08       53.73
3. No spouse or unmarried partner prese |     63,377       46.27      100.00
—————————————-+———————————–
Total |    136,960      100.00

You would think that this shouldn’t change the results much.  Except it does.

. ttest  bls_hhact_hwork if spousepres==1 | spousepres==2, by(hh_child)

Two-sample t test with equal variances
——————————————————————————
Group |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
———+——————————————————————–

0. No |   30366    40.27857    .4782439    83.33803    39.34119    41.21595
1. Yes |   43217     46.9683    .4089027    85.00556    46.16684    47.76976
———+——————————————————————–
combined |   73583     44.2076    .3110836    84.38513    43.59788    44.81733
———+——————————————————————–
diff |           -6.689731    .6314013               -7.927276   -5.452187
——————————————————————————
diff = mean(0. No) – mean(1. Yes)                             t = -10.5951
Ho: diff = 0                                     degrees of freedom =    73581

Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
Pr(T < t) = 0.0000         Pr(|T| > |t|) = 0.0000          Pr(T > t) = 1.0000

Having a child when you have a partner in the house increases housework by 6.7 min.

For cohabiters it’s an increase of 12 min!

. ttest  bls_hhact_hwork if spousepres==2, by(hh_child)

Two-sample t test with equal variances
——————————————————————————
Group |     Obs        Mean    Std. Err.   Std. Dev.   [95% Conf. Interval]
———+——————————————————————–

0. No |    2112    33.51657    1.586618    72.91542    30.40507    36.62807
1. Yes |    2112    46.34943    1.895052    87.08996    42.63307     50.0658
———+——————————————————————–
combined |    4224      39.933    1.239569    80.56248    37.50279    42.36321
———+——————————————————————–
diff |           -12.83286    2.471554               -17.67841   -7.987314
——————————————————————————
diff = mean(0. No) – mean(1. Yes)                             t =  -5.1922
Ho: diff = 0                                     degrees of freedom =     4222

Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
Pr(T < t) = 0.0000         Pr(|T| > |t|) = 0.0000          Pr(T > t) = 1.0000

Married people spouse present:

Housework_min = 14.08 + 51.56*Female + 4.35*hh_child + 2.43*(Femalehh_child)

Everything significant at the 5% level.  (Results are similar with weighting)

Cohabiters:
Housework_min = 19.42 + 29.04*Female + 2.01*hh_child + 14.56*(Femalehh_child)
(results with weighting are pretty similar, with an even bigger interactive effect)

hh_child is not significant.  Note how much less housework cohabiting women do compared to married women!  (29.04 vs 51.56)  And look how much bigger that interaction of having a child is for cohabiting women– a child only adds 2.43 min (plus the 4.35 main effect that it adds to both parents) to married women, but it adds a full 14.56 minutes to cohabiting women (18.5 minutes in the weighted regression).  The story here is that cohabiters did less work and then were forced to be more traditional once a baby arrived.  With married women we’re probably seeing a lot of housewives increasing that female coeff.  There could also be differences in hiring out help between people who cohabit and people who are in more traditional marriages.  Or in how big the house/apartment is.  There are a lot of controls that could be put into these regressions (age, labor force status, etc.) if one wanted to try to get at causation instead of just the relationships.

Grumpy nation, how does this square with your experience, if applicable? And isn’t Stata awesome?