I decided to write the paper as a blog post and that's how I'm turning it in (I'm not doing two versions, one for public consumption and one for "publication"). Admittedly, that's pushing things a bit. People don't blog about statistical analysis and final papers aren't usually blog posts. But you know, I firmly believe that anything can be made interesting to a wider audience than the narrowly defined scholarly research audience that we typically prepare our work for. The writing just has to tell a story that's reasonably interesting to real people in a language they can understand -- mainly that means avoiding throwing unusual terms at the reader without explaining what they mean.
So I have to explain certain of the parentheticals here, the ones like this: (b=.759; t=-4.96; p=.000). This is statistics codespeak. It truly is a foreign language, more difficult to interpret even than legalese in the sense that normal people can't understand what it means. Even if you translate it from symbols and numbers to plain English, it still does not make ordinary sense (the slope is .759; the t-score is 4.96; the p-value is .000). No, it takes a lot more words to explain it than that. But, that's what the text is all about. It explains what the code shows without specific reference to the code. The parentheticals are for the statistically-inclined only, those for whom the raw data speaks volumes of "backup" for the statements in the text. They are similar to the citations in legal writing. So feel free to ignore them if you don't want to delve into the supporting authority. I couldn't leave them out. They matter, just not to everyone.
Writing so that unusual terms are explained is good writing anyway, for scholars or for anyone. The difference between scholar-speak and blog post mainly comes down to style. What is it about scholar-style that so quickly brings on the glaze-over for most people? Maybe I'll write about that some day. And, I am sure I probably don't have the style right. But I'll keep working on that. In the mean time, here's the stats paper:
Will retirement slow your personal growth?
Shouldn’t it be the other way around?
So, a funny thing happened the other day as I was looking over some data from the 1993-1994 wave of the Wisconsin Longitudinal Study (WLS), you know, the study where researchers repeatedly ask a random sample of more than 10,000 Wisconsin high school grads, their parents, and later their siblings or widows, tons of questions about just about everything. I found a negative relationship between retirement and personal growth! Analyzing the respondents’ answers using a mathematical “model,” you can roughly predict a person’s responses regarding personal growth if you know his or her retirement status: those who are retired will have lower scores for personal growth, on average .76 lower, than those who are not retired. This is pretty depressing. I just retired two years ago. So, shall I just say, ‘so long’ to personal growth? Well maybe I don’t need to worry. These differences between growth scores for the retired and the employed are not alarmingly large (even though the difference is statistically significant), and folks responding to the surveys administered by the WLS graduated from high school in 1957, which means they were born roughly between 1939 and 1941. I, on the other hand, am a boomer, a member of the generation born between 1946 and 1964, estimated at near 80 million. We’ll be retiring over the next 10 to 20 years. As we have in every other phase of our lives, we will affect the understanding of this phenomenon if for no other reason than that our numbers challenge social structures and facilities designed to handle much smaller volumes of individuals at a given time. But more fundamentally, we’ll likely challenge and ultimately change even the very definition of what it means to retire (AFP, 2007).
Leonard Steinhorn, an American University professor and author of "The Greater Generation: In Defense of the Baby Boom Legacy," says the generation often wrongly maligned as latte-sipping Yuppies has transformed most of American society.As reported at a world gathering of statisticians in 2007, a Canadian researcher reached a similar conclusion about the likely change in definitions for retirement and noted a lack of statistics about this phenomenon (Bowlby, 2007):
He wrote that boomers have led or sustained most of "the great citizen movements that have advanced American values and freedoms -- the environmental movement, the consumer movement, the women's movement, the civil rights movement, the diversity movement, the human rights movement, the openness in government movement."
He told AFP he expects this transformation to continue as boomers age. "It's not going to be a generation that's going to go off to the golf courses and do nothing."
Even though this wave [of boomers] will have significant labour market consequences over the next 20 years, no regular statistics are produced on the retired population.
There are some reasons for this. Only recently has the need for retirement data grown. Secondly, the concept of retirement is fuzzy, to say the least. Retirement can mean different things to different people, and measuring it is difficult for national statistical organizations. Having an international standard would assist in deciding what data or range of data should be produced.
It seems implausible to me that a generation that likes to characterize itself as active, engaged, lifelong learners will just let growth slide upon retirement. There really must be more to this story. Maybe there are other factors that explain the connection between retirement and reduced personal growth for the generation that precedes the boomers into retirement, factors that can help predict whether boomers’ growth will stagnate.
To find out, I examined the interactions of these two variables with three others from the WLS to try to explain reduced personal growth at retirement: I investigated how the negative relationship between retirement and growth is affected by one’s sense of purpose; whether the relationship depends on gender; and how educational levels affect post-retirement growth.
Bowlby may be concerned about the lack of research on retirement and the labor force, but there is definitely plenty of research on retirement and health, summarized by Phyllis Moen in her 1996 article, “A Life Course Perspective on Retirement, Gender and Well-being.” As she reports (p. 132), “there is some evidence that being retired affects psychological symptoms” (citations omitted). She also notes that “from the perspective of social integration, individuals are better off both physically and psychologically when they have a greater number of roles, leading to a sense of purpose, identity, and community (citations omitted). Thus, older individuals who delay retirement or take on subsequent paid work or unpaid volunteer work following their retirement from their career jobs would be better off than those who retire from work without compensating roles and relationships” (p, 133). Hardly surprising, is it?
But Moen also noted significant holes in the research, areas ripe for future study. She cites as examples “[t]he effects of gender … as a variable moderating the retirement and well-being link” (p. 132) and “research on the employment-health linkage [that] consider[s] postretirement employment, and … the effects of retirement on women's life patterns and health (citation omitted)” (p. 139).
“Moreover,” she says, “scholars are only beginning to investigate the health consequences of involvement in unpaid volunteer activities (citations omitted)” (p. 139).
Clearly, there is much we do not know about how boomers will retire, what they’ll make of it, and how retirement will affect their psychological health and well-being.
As I noted, the WLS surveys more than 10,000 randomly selected 1957 high school graduates, or those able to provide information about them (parents, spouses, siblings), at intervals ranging from one to 15 years. The survey questions are designed to provide information about participants’ “relationships, family functioning, physical and mental health and well-being, and morbidity and mortality from late adolescence through middle age” as well as “social background, youthful aspirations, schooling, military service, labor market experiences, family characteristics and events, social participation, psychological characteristics, and retirement” (WLS, 2008).
I want to know more about personal growth. The WLS measures personal growth with the following questions:
To what extent do you agree that
- you are not interested in activities that will expand your horizons?
- you have the sense that you have developed a lot as a person over time?
- when you think about it you haven't really improved much as a person over the years?
- you think it is important to have new experiences that challenge how you think about yourself and the world?
- you don't want to try new ways of doing things -- your life is fine the way it is?
- you do not enjoy being in new situations that require you to change your old familiar ways of doing things?
- there is truth to the saying you can't teach an old dog new tricks?Respondents chose their answers from a 6-item scale. They strongly, moderately or slightly agreed or they slightly, moderately or strongly disagreed. Total scores ranged from 1 to 42. The average for respondents’ scores was 32.69.
Initially I wanted to know how retirement affected personal growth. WLS respondents indicated whether they were retired by choosing among five possibilities: partly retired, completely retired, working and not retired, not working and not retired, and don’t know. These five had been collapsed into two categories of retired or not retired for the dataset that I examined. Only 18% of respondents indicated that they were retired.
Other variables that may explain the relationship between retirement and personal growth
Theoretically, retirement could be a time of exploration, of branching out, of trying new things and taking on projects one never had time for before. The WLS data suggests that may be wishful thinking. When I learned that, far from spurring personal growth, retirement seemed to squelch it, I wanted to know more. I decided to look at sense of purpose in life, gender, and educational attainment to see what I could learn about their possible roles in this negative trend in growth after retirement.
Purpose. Sense of purpose is measured with the same 6-item scale described above for personal growth, but, of course, the questions are different:
To what extent do you agree that
- you enjoy making plans for the future and working to make them a reality?
- your daily activities often seem trivial and unimportant to you?
- you are an active person in carrying out the plans you set for yourself?
- you tend to focus on the present, because the future nearly always brings you problems?
- you don't have a good sense of what it is you are trying to accomplish in life?
- you sometimes feel as if you've done all there is to do in life?
- you used to set goals for yourself, but that now seems like a waste of time?
Gender. Gender is, of course, either male or female. There were slightly more women than men respondents.
Educational attainment. Education is measured in years of attainment. Respondents completing 12 years (a rough estimate for those graduating high school) constituted 54.1% of the total. Another 13.5% completed 16 years (likely graduating college) and 13.2% finished between 17 and 21 years of formal education. The overall mean number of years completed was 13.6.
Analytic Plan and Results
So, my purpose was simply to explore whether that negative association between retirement and personal growth persists if we take other things into account, or “control” for other variables. The analysis proceeded through several steps as I’ll explain in more detail below. First, of course, I figured out exactly what the relationship between retirement and personal growth was by looking at correlations between a number of the variables from the Study. These correlations show which variables change in associated patterns, either negative (if one goes up the other goes down) or positive (if one goes up the other goes up or if one goes down the other goes down), and how strong the associations are (how likely or unlikely they are to occur just by chance). These associational clues enabled me to think about which of the many variables might have effects on the relationship between retirement and growth. I also examined detailed descriptions of many of the survey responses. For example, I looked at how many people were men, how many were women, how many were retired, how many completed 12, 16 and more years of education, what were average scores for growth and purpose, were growth scores on average higher or lower for the retired, for women, for men, etc. Finally I chose three variables to examine more closely. I looked at what happened if I took the respondents’ sense of purpose in life into account. Next, I examined how the relationship changed depending on gender. And then I evaluated how educational achievements affected the relationship.
Analyzing the Main Relationship: The Bivariate Analysis
As I reported in the introduction above, I found that those who were retired had scores for personal growth .76 lower than those who were not retired. For example, the average personal growth score for those who were not retired was 32.96; the average for those who were retired was 32.12. I discovered the dimensions of this relationship by analyzing all respondents’ scores for the two variables, retired and growth, to see how much personal growth scores changed, on average, with a one-unit change (that is, from 0 to 1, from not retired to retired) in retirement status. In technical terms, I conducted an ordinary least squares bivariate analysis, or OLS regression, on the two variables (bivariate!). This analysis told me that they changed together, and that the way they changed together was extremely unlikely to happen just by chance (that is, it was highly statistically significant (b=.759; t=-4.96; p=.000). But it also showed that the relationship was not likely to be of much practical importance because retirement only explained ¼ of 1% of the variability in respondents’ personal growth scores, and as indicated above, there is only an average difference of .76 between the growth scores for the retired and the employed. Clearly, there were a lot of other things that explained personal growth in addition to retirement. Nevertheless, I still wanted to find out more about this negative association.
Considering Additional Variables: The Multivariate Analyses
Controlling for purpose
I ran the same analysis (OLS regression) again, but this time I added purpose in life to my model for predicting growth scores (“controlled for” purpose). This allowed me, in effect, to look at the relationship between retirement and growth for groups of the respondents who all had the same scores for purpose in life. I found that the negative relationship I first observed practically disappeared -- it diminished by 99.5%! The association between the two, controlled for purpose, could easily occur by random chance (that is, it lost all statistical significance (b=.004; t=-.03; p=.97)). Interestingly, retirement and sense of purpose together explained almost 45% of the variability in personal growth, an increase over what retirement explains by itself of over 180%.
It seemed a bit too convenient that the drop in personal growth might not really be associated with retirement at all, but with a loss of one’s sense of purpose. Voila! End of inquiry! But this wasn’t very satisfying. It seemed like I still faced the same question –why? On the other hand, this did make some sense: maybe retirement leaves people without a sense of purpose. “I am [my career], so when that is over, what am I after that, and what’s my purpose for getting up every day?”
I checked with my instructor. She was not surprised at all by my finding. She pointed out that many people believe that one’s sense of purpose and one’s personal growth reflect the same underlying dimension of psychological well-being. They might be “interchangeable,” so to speak.
I looked at the questions the WLS uses to evaluate these two variables (that’s why I repeated them above, so you could see them too). I couldn’t really tell for sure if they were getting at the same thing, though it seems possible, so I tested that idea by reversing the analyses: If I were measuring the relationship between retirement and purpose in life, would controlling for growth explain away the (presumed negative) association?
Well, there was a negative relationship but growth didn’t completely explain it away. Controlling for growth reduced the relationship between retirement and purpose by 46% (b=-1.17; b=-.63), but it remained statistically significant (t=-5.4; p=.000). So purpose and growth were not exactly interchangeable, but it does appear that they are very closely related. As one goes up (or down) the other changes in the same direction at about 2/3 the value of the first (Pearson r = .68). Perhaps retirement has a more direct effect on purpose, and only an indirect effect on growth, as suggested above, but the strong association of these two makes them a poor choice for a beginning statistics student, so I’ll move on.
Controlling for gender
Next I turned to gender to see what effect it might have on the relationship I had observed between retirement and growth. But this time, controlling for gender in the same way I had earlier controlled for purpose (running a multivariate OLS regression) nudged the negative effect of retirement on growth up 9% (b=-.859; t=-5.62; p=.000). This finding, along with the fact that employed women’s scores for growth were, on average, 1.05 points higher than employed men’s scores, suggests that gender suppresses the effect of retirement on growth. The effect is too small to conclude that gender can explain away the effect of retirement on growth, but it does show that men and women experience personal growth and retirement differently.
Controlling for educational attainment
Next I asked whether the relationship I had observed might be affected by the level of formal education completed by the respondent. Controlling for education (running a multivariate OLS regression) resulted in a 42% reduction in the negative effect of retirement on growth (b=-.76; b=-.44). The relationship was still negative and still statistically significant (t=-2.9; p=.004), but not nearly so bad as it first seemed. Of course, education was positively related to growth: for every additional year of education, the growth scores for the employed rose .5, and this correlation was highly significant (t=21.38), but who would be surprised by that? It does suggest that the main relationship may be spurious in that almost half the effect of retirement on growth is really due to education. Still, we have to keep in mind that education and retirement together account for only about 4.7% of the variability in growth scores.
I decided to have a closer look at the effect of education on the relationship between retirement and growth. As I had earlier noticed in the correlations, higher levels of education are associated with lower values for retirement (that is, the more educated tend not to be retired) and higher values for growth (r=-.1; p=.0000; r=.22; p=.0000), so it looks like the three-way relationship is probably complex.
I created a new variable that would allow me to compare the growth scores for retired and employed who had a high school diploma (12 years of education), with scores for those who had a college degree (16 years of education), and with scores for those who had considerable post-graduate education (20 years) – an “interaction term” that would show how retirement and education are likely to interact in their effects on growth.
By running another OLS regression including the education and retirement variables from the WLS Study plus this new interaction term, I found that, indeed, the effect of retirement on growth was significantly different depending on how many years of education one has (b=.16; t=2.42; p=.015). Retirement was still significantly negatively related to growth for those with average education, but by considerably less than before taking education and its interaction effects into account (b=-.39; t=-2.59; p=.01): Growth scores for those with average education were only .39 smaller for the retired (recall that average education is 13.6 years). The effect of education on growth for those who are employed was still highly significantly and positively related, as one would expect (b=.49; t=18.55; p=.000): One additional year of education will raise growth scores for the employed by about .49 points.
Now let’s look at growth scores for the employed and the retired who had less or more than “average” years of education.
A post-graduate degree will raise growth scores for the retired slightly more than 6 points above scores for those who only finish high school. That degree will only boost scores for the employed by just under 4 points above the scores of the high school grads.
But the employed who only graduated high school start out with higher growth scores than their retired respondents. At 20 years of education, the tables have turned and the growth scores of the employed lag behind the retired! And as you can see, at 16 years of education, retirement status is irrelevant -- whether you are retired or working, your growth score will likely be the same.
The gaps between the employed and the retired at the extremes of education I examined are not large, but there is a bit of a difference between the high school grad gap and the advanced degree gap. The gap at 12 years is almost 2 points (1.7); at 20 years, the gap is just a little more than half a point (.6).
In summary, growth scores for the retired really only lag behind those of the employed for those with less than a college degree. Thereafter, additional years of education are correlated with higher growth scores for the retired than for the employed.
Life changes when we retire. Retirement can be a time of freedom from the exigencies of earning a living and raising a family, giving us time for renewed curiosity and exploration; or it can be a time of uselessness, boredom and resignation. Of course, it’s more likely to be something in between. Either way, some characteristics of our lives before retirement might give us clues as to how we will likely behave or feel after we retire.
It appears that the better educated among us are not so eager to retire, and once they do, they’ll be more likely than their less educated peers to continue to grow in retirement. Women may have a slight edge over men in the degree to which they continue to grow after retirement, but education is more helpful than gender in explaining what might keep us curious and learning after we retire. It would seem that the same things that prompt some of us to pursue more years of formal education before we retire may continue to drive us to learn and grow afterwards.
Limitations and Future Directions
The 1993-1994 wave of the Study surveyed these Wisconsin graduates in their mid- to late-50’s – not quite retirement age by most standards, though according to Bowlby, at least in Canada, retirement age had been trending lower through the last half of the 20th Century (para. 6). He observes that the trend has begun to turn around in the fifteen years since. The Study data show that only 18% of respondents identified themselves as retired. Moen’s summary of research conducted during the same part of the 20th Century indicates that early retirement is often caused by declines in health; early retirement does not itself cause declines in health). This suggests that it’s at least possible that the leading edge of retirees from this generation was in retirement at that time because of poor health, which could certainly affect responses to questions about personal growth and sense of purpose as much or more than retirement status. And indeed, the average response for self-rated health for the group of retirees is lower than the average response for those who were not retired (means of 3.96 and 4.18, respectively). This difference in scores is only .22, but it is highly statistically significant (t=12.8; p=.000). It certainly suggests that we ought to review the findings about retirement and growth based on the 1993 – 1994 wave when a larger percentage of participants has retired.
AFP: US braces for baby boom retirement wave. (2007, December 24). Agence France-Presse. Retrieved November 23, 2008, from http://afp.google.com/article/ALeqM5g-W4yeMsbNpknzofffLTFxOgKI2A.
Bowlby, G. (2007, February). Perspectives on Labour and Income - Defining retirement. Retrieved November 24, 2008, from http://www.statcan.gc.ca/pub/75-001-x/10207/9584-eng.htm.
Moen, Phyllis. (1996). A Life-Course Perspective on Retirement, Gender and Well-Being, Journal of Occupational Health Psychology, Vol. 1, No. 2, pp. 131-144.
Wisconsin Longitudinal Study (WLS). (2008, November). University of Wisconsin-Madison. Retrieved November 26, 2008, from http://www.ssc.wisc.edu/wlsresearch/.