In the late 1960s in the United States, when divorce was first beginning to become common, there was a clear answer: children had a strong, statistically significant stabilizing effect on marriage, at least for the first few children and at least while the children were young. In a 1977 paper reporting on data current as of the late 1960s, Becker et al. found that the presence of a child under the age of 6 by the fifteenth month of marriage reduced the probability of divorce in the first five years of marriage by about 25% (removing about one percentage point of an approximately four percent risk of divorce), and had an even greater effect in the second five years of marriage. (A second child, however, reduced the risk of divorce less than the first - and a fifth child seemed to actually increase the risk of divorce. In addition, the risk of divorce was increased if the wife was pregnant prior to the marriage.)
That children should be stabilizing to a marriage makes perfect evolutionary sense and hardly needs explanation. Marriage is a union for the (evolutionary) purpose of supporting common genetic interests; children are exactly the "marital capital" that would be expected to make both parents more willing to invest in the partnership. The presence of children also makes divorce less attractive to the members of the couple, as each would have a harder time finding a new mate given the financial, social, and physical toll children take on one's subsequent mate value.
That NOT having children should destabilize a union would make excellent evolutionary sense to the extent that sterile unions were a serious risk under the environments of evolutionary adaptedness. To the extent that sterility was a risk for couples in EEAs, whether through genetic incompatibility, infectious disease, or other reasons, there would be major fertility benefits to exiting a union if it failed to produce children (at least for the fertile partner). If sterility were rare in the EEAs, a behavioral adaptation facilitating the abandonment of a childless union might not be adaptive.
So in the 1960s, when legal marriage was still somewhat synonymous with lifetime partnership, having one or two young children did, in fact, contribute to marital success. However, this has not been the case with cohabitation relationships; having children actually strongly destabilizes a cohabitation relationship.
I have argued that because humans evolved to rely on cultural patterns and coercion to enforce marital partnerships, marriage no longer really exists; all that is left to us is cohabitation relationships under the false name of marriage. If this is true, we would expect the presence of children to begin destabilizing (what is now legally recognized as) marriage, rather than stabilizing marriage, in proportion with the reduction of coercion and other cultural patterns that support marriage. Sadly, in a 2001 study from the UK by Chan & Halpin, this is exactly what was found to have occurred. The study examined multiple sources of data from the 1950s through the 1990s and found that while the presence of a small child had a stabilizing effect on marriage in the 1950s, the effect began to change sign in the 1980s and was large and negative by the 1990s:
For those who got married in the 1950s, each additional child was associated with a 16% reduction in divorce risk. In contrast, for the 1990s marriage cohort, each additional child is associated with a 37% increase in divorce risk. [Emphasis mine.]
Even young children - those most stabilizing in the Becker paper - were found to be destabilizing by the 1990s. Chan & Halpin found no effect for the gender of children, but note that some authors have found that sons are less destabilizing than daughters. The cause the authors found most likely for the shift was income inequality, as income inequality increased drastically in the UK during the 1980s and 1990s and the destabilizing effect was most prominent for low-income families.
In modern life, the birth of a child is associated with a sudden decrease in marital satisfaction that generally persists throughout the relationship. Similarly situated couples who do not have children experience a more gradual decline in marital satisfaction. How might this observation be explained in terms of the effect of children on marital success?
One hypothesis is that children have always caused marital satisfaction to plummet, but that patterns and coercion were strong enough to cause a troubled couple to remain together. A second hypothesis might be that the sudden drop in marital satisfaction with the birth of a child is new, and reflects the loss of happiness-maintaining patterns that would, in EEA or even pre-1980s conditions, prevent the drop in marital satisfaction that is now observed shortly after the birth of a first child. Those investigating children and happiness in the 1970s and 1980s found a decrease in happiness with children (see Baumeister's Meanings of Life, Appendix B: The Parenthood Paradox), including a decrease in marital satisfaction, but it is not clear if this change predates the modern loss of marriage patterns.
Whatever the explanation, marriage as it currently exists is not providing the kind of stable, reliable union of genetic interests that can survive the stress of the birth of a child - and poor children are those most harmed.