The Research Page of Christopher Steven Marcum

I am a mathematical sociologist working in the area of health, aging, and social networks. As a generalist, I have published work in a variety of areas including organizational sociology, behavioral medicine, gerontology, and public policy. I also have expertise in large dataset management and analysis.

A current version of my cv can be found here.

Given the latency between updates on this website, please see these up-to-date external sites for my recent publications:

Publications

All of my publications can be found on my dropbox. For a list, see my cv.

Research at NIH

I am a faculty member of the Social and Behavioral Research Branch (SBRB) at NHGRI. I work very closely with Laura Koehly as a staff scientist and methodologist in the Social Network Methods Section. My research agenda is broadly situated within the context of social networks and health. My current projects support and advance our section’s mission and include papers that: 1) demonstrate the utility, validity, and reliability, of a novel approach for multiple-item measurement of social network data; 2) use network inference methods to identify key actors in caregiving networks to people with Alzheimer’s disease and related dementia (among other genetic diseases) based on multiple-informant accounts; 3) develop methods to learn about relationships among Mexican-origin youth living in the same community from a combination of genetic admixture, social, and geographic data; 4) model the inter-generational communication patterns within Mexican-origin families as they discuss health concerns; and, 5) model dynamic social health behaviors unfolding in time on a network of actors. The substance of this agenda taps my training in gerontology by addressing health-related issues across different phases in the life course (i.e., from youth, to middle adulthood, and finally to old-age).

I also act as a mentor to SBRB trainees and teach a seminar on networks and statistics in the branch. Specifically, I supervise the dissertation of OxCam fellow Mr. Jeffrey Lienert (with L. Koehly and F. Reed-Tsochas) and supervise the post-doctoral training of Dr. Jielu Lin (Ph.D., Case Western 2014).

Research at RAND

I collaborated with RAND faculty during my post-doctoral work. One project with physicist Raffaelle Vardavas employees an inductive reasoning model of influenza vaccination decisions as affected by influenza diffusion on large-scale social contact networks. My role is two-fold: parameterizing the IR model using sociological theory and empirical data and constructing realistic models of large-scale contact networks. The goal of the project is to advance vaccination policy (i.e., how to minimize effective coverage in the population) above-and-beyond recommendations arising from agent-based research.

Another project involved developing models of daily social behavior across the life course. For this work, I am used the sequence data from the American Time Use Survey to understand how age differences in the process of day-to-day living arise. This project attempts to shed new light on the micro-macro link between very local activity sequence transitions and age-associated declines in daily social contact.

Research for Fun

Sometimes, I do research for fun. Here are a few examples:

Dissertation Research

My dissertation research focuses on structural explanations of age-differences in daily social interaction. I use the American Time Use Survey to model patterns of who interacts with whom, doing what, at what time, and how long, in an average day. This "snapshot" approach to social network research allows for in-depth analysis of how personal networks are activated in an everyday context. My emphasis for this research is on how American's personal networks differ across the life course, and especially by age group. My dissertation research sheds light on the processes by which age differences in social behavior arise.

A couple of illustrations from my dissertation research are provided below.

Most Common Daily Activity Transitions Sequences
Since the American Time Use Survey disregards simultaneous behavior (i.e., multi-tasking), we can construct sequences of behavior as transitions from stopping one type of activity and starting another. This figure boils down the most common transitions into a relational (path) diagram. The red nodes represent classes of activities and the directed edges represent transitions from one activity to another. Here, an edge from one node to another represents the cessation of the exit activity and commencement of the entrance activity. For example, the arrow from Leisure to Sleeping should be read as "Stopped Leisure and Started Sleeping." Nodes are scaled by the relative frequency of the activity spell (but not their average duration). Determination of which transitions to plot was made by using hierarchical sequence analysis program Sequitur on the entire ATUS dataset. Sequences that occurred more than 5000 times are plotted. The five isolates are valid activities but did not appear in enough sub-sequences. Plotting was accomplished using the plot.network() in Butts's network package for R.

Comparing Age Differences in Diurnal Activity Sequences
This figure illustrates differences in the activity sequences from 2000 randomly selected individuals over the course of a single day in two age groups: 25-34 year olds to those age 75 and older. Each row in the figure represents one individual and each colored bar represents an activity spell. The matrix that the figure represents is sorted by Hamming distance, with more similarly clustered activity sequences closer together and more distant ones further apart. The figure clearly shows that the younger group has greater activity heterogeneity in both the types of activities and the order in which they are done, than the older group. In particular, and as expected, domestic and work production activities are replaced by sleeping and leisure activities moving from 25-34 to the 75 and up groups.

Social Time with Select Relations (in Hours) by Age
The average number of hours spent with different types of people and time spent alone is plotted against age, with 95% probability bands, in this figure. The figure illustrates how older people spend less time with others overall, and more time alone. Time spent with children declines after middle-adult and continues on a downward trajectory through old-age, while time spent with other relatives (non-spousal, non-children relatives) flatlines at about 2 hours in an average day by the late 20s. The irregular M-shaped curve of spousal time is so-shaped due to an increase in divorces in middle-adulthood followed by widowhood in very old age. Clearly, there is great variability in the amount of time people spend with different types of relations across the life course.

Interaction between Age and Subjective Health Status on Time Spent Alone
Both old age and poor health are states that are associated with feelings of loneliness, increased risk of social isolation, and in this case, spending time alone. Aging and poor health are also positively associated and we expect that the age effect of spending time alone will vary by health status. The figure illustrates the interaction between age and subjective health status on the expected number of hours spent alone. The dependent variable counts time that could have been spent with other people, and thus excludes certain personal activities and sleeping (and working activities because of ATUS coding protocols). Younger people in poor health are more different from younger people in good health than their older counterparts, in terms of spending time alone. Indeed, the difference in hours spent alone between being in bad health and being in good health declines with age. This suggests that the marginal effect of aging on increased time spent alone out-paces the effect of declining health status associated with aging. This figure plots the gross interaction effects and does not control for other exogenous factors that may shape the relationship, such as widowhood or retirement statuses.

Interaction between Age and Employment & Marital Statuses on Time Spent with Non-Kin
Work and family life often shape who we interact with. Two of the ways in which the work/family dynamic influence social interaction is through the different social contexts associated with various marital and employment statuses. This is especially salient for shaping how much time we spend with people outside one's household. Of course, people of different ages have different propensities to fall into one or another marital and employment statuses. We would expect that the effect of marital status and employment status on social time with non-kin to vary with age. The first panel plots the average number of hours of social time spent with Non-Kin relations by the age by employment status interaction between persons who are employed (the baseline), unemployed, retired, and not in the labor force for other reasons (i.e., education or disability). The second panel plots the results for the age by marital status interaction between persons who are currently married (or living as married, which is the baseline), never married, divorced (or separated), and widowed. These lines are derived from the posterior means of the Bayesian multivariate linear regression of social time spent with various types of relations and spent alone on covariates (MCMC draws n=1000). The big take-away from these figures is that age interacts with marital and employment status in very different ways. The differences between marital statuses at younger ages are greater than they are at older ages, suggesting that the effect of aging reduces the effect of marital status on the ability to spend time with non-kin. This is contrasted by the interaction between age and employment status, which shows an opposite pattern. The differences between employment statuses are smaller at younger ages than at older ages. By the interocular percussion test, work and family life compete in opposing directions to constrain interaction with friends, colleagues, and other non-kin across the life course.

Note: The errors from the regressions were allowed to correlate, which accounts for the clustering of relationship types (i.e., within a family) and the negative relationship between spending time alone with all other types of social time.

Seasonality Effects on Social Time by Day of the Month and Month
Averaging across all years, this time series plots the copresence volume in the American Time Use dataset by days nested in months for all types of relations analyzed in my dissertation. The dependent variable is the sum (volume) of person hours spent copresent with others in the dataset. Weekly seasonality effects with no obvious trend are apparent in each month, with weekly peaks corresponding to weekend days and lulls corresponding to week days. Some peaks and outliers in the data appear to be associated with major US Holidays, demonstrating the importance of discrete events in interrupting both seasonality effects, in particular, and the probability of interaction more generally.

Other Research

In addition to my dissertation, I have a number of other research interests. Briefly, I have worked on projects related to organizational collaboration networks, population-level attitudinal change, and statistical research methods.

Old Posters

Here are few of the scientific posters I have presented at conferences in graduate school.
Attachments
File Last modified Size
paa2009.pdf 2010-07-30 00:48 137Kb
hazards2010.pdf 2010-07-30 00:51 260Kb
muri2009.pdf 2010-07-30 00:51 704Kb
paa2010.pdf 2010-07-30 00:51 789Kb
muri2010.pdf 2010-12-11 19:54 1011Kb


Resources

There are a number of research resources I use to stay productive. These include: R for statistical computing; Latex for document typesetting; and subversion and rsync for archiving and organizing my digital files. For the most part, I try to maintain a library of software that is open-source and free.
Valid XHTML :: Valid CSS: :: Powered by WikkaWiki