(This page will show up differently on different web browsers, so be sure to scroll down the page to see all that is available.)

JUNE 2006 -- New article published in academic journal. 
See list of papers from project (below).

Alan Reifman, Principal Investigator
E-mail:  alan.reifman@ttu.edu
Web:  http://www.depts.ttu.edu/hdfs/reifman.php

Wendy Watson, Research Assistant, 1999-2000
Andrea McCourt, Research Assistant, 2000-2001

Department of Human Development and Family Studies
College of Human Sciences -- Texas Tech University

• Funded by Grant No. R03 AA12021 from the National Institute on Alcohol Abuse and Alcoholism to Alan Reifman.
  
• See article on the project in the College of Human Sciences' Viewpoints magazine (Spring/Summer 2000).

Please scroll down or click on the headings immediately below to get to your desired section:

MAJOR FINDINGS
LIST OF PAPERS FROM PROJECT
RESEARCH METHODOLOGY
LINKS TO OTHER ALCOHOL-RELATED SITES ON THE WEB

MAJOR FINDINGS (SO FAR!)

Nationally, many observers have been concerned over the degree of heavy episodic or "binge" drinking that takes place across this country's college campuses.  The landmark College Alcohol Study, conducted on a national sample by the Harvard School of Public Health, defined a binge episode as five or more drinks in a row for men, and four or more for women.  A "binge drinker" was defined as someone who had at least one of these episodes in the last two weeks.  The Harvard study found that 44% of American college students were binge drinkers, according to these criteria (Wechsler and colleagues, 1994, Journal of the American Medical Association).  

The present study, which surveyed only students at Texas Tech, used the same definition of binge drinking.  As shown immediately below, the surveyed Texas Tech students engaged in binge drinking at a slightly lower rate than the national average.  

There are a couple of factors that must be kept in mind, however.  First, the above graph included only students who participated in all three measurements of the project.  If we look at all 273 students who answered the binge-drinking measure in the Fall 1999 survey (regardless of whether they stayed in the study over time), the rate of binge drinking at that time would have been 45% (or .45), a much closer match to the national rate.  Second, even though the Texas Tech project surveyed students only through their sophomore year and Harvard's national study surveyed students representing all four years of college, this should not matter much, because the Harvard study found no differences in binging according to year of school.

Our Texas Tech study was concerned with much more than just documenting rates of heavy drinking, of course.  The main focus was examining the role of students' social networks in the students' own drinking.  Numerous studies have shown that heavy-drinking individuals also tend to hang out with heavy-drinking peers.  But which way does the direction of causation run?  One possibility, which we refer to as social influence, is that other people's drinking affects the focal individual's drinking, through processes such as peer pressure or modeling (i.e., imitation).  With social influence, the direction of causation is from the network to the individual.  Another possibility, however, is of a process called selection.  Here, an individual who is already a heavy drinker seeks out (or "selects") other heavy drinkers and forms a network with them (or joins a pre-existing one).  With selection, the direction of causation is from the individual to the network.  The following figure attempts to illustrate social influence and selection (added 3/19/07):

A major aim of our study was to determine whether social influence, selection, or both were in operation during our respondents' first two years of college.  A research design known as a "panel study" was used; in such a design, the same respondents are surveyed over time.  As noted above, students completed surveys at three points in time:  first semester of the freshperson year (known as "wave 1"), second semester of that year (wave 2), and first semester of the sophomore year (wave 3).  Among other things, students reported on their own drinking and the drinking of their network members at each wave.  The basic idea is that we can look for whether individuals with heavy-drinking networks at an earlier wave go on to increase their own drinking at a later wave (social influence) or individuals who are heavy drinkers at an earlier wave gravitate to a heavier-drinking network at a later wave (selection).  Although causality cannot be established with the same certainty as in a tightly controlled laboratory experiment, the panel nature of the research design can show temporal precedence (i.e., what comes before what).  Click here for further information on the logic of this research design.

Evidence was found for both social influence and selection.  Students with a heavy-drinking network at wave 1 increased their own drinking by wave 2, and the same occurred between waves 2 and 3, supporting the idea of social influence.  Also, however, students who were already heavy drinkers at wave 1 gravitated to heavier drinking networks at wave 2, supporting the idea of selection (this latter finding did not replicate to the interval between waves 2 and 3).

We gathered a lot of data on many topics besides just social networks, so we hope to be generating ideas and research papers for a long time to come!  A list of scholarly papers to come out of the project (which we hope to be able to update frequently) can be located by scrolling down further on this webpage.  Thank you for your interest in our project.

LIST OF PAPERS FROM PROJECT (SO FAR!)
You can e-mail Dr. Reifman for copies of papers that we've not put on the web.

Reifman, A., Watson, W.K., & McCourt, A. (2006). Social networks and college drinking: Probing processes of social influence and selection. Personality and Social Psychology Bulletin, 32, 820-832.

Reifman, A., & Watson, W.K. (2003). Binge drinking during the first semester of college: Continuation and desistance from high school patterns.  Journal of American College Health, 52, 73-81.

McCourt, A., & Reifman, A. (2002, August). Social connectedness and mental health in college students. American Psychological Association. Chicago, IL.  (Click here for copy of paper.)

Rethinam, V., & Reifman, A. (2002, April). Does closeness to parents and peers regulate college students’ modeling of their drinking?  Society for Research on Adolescence. New Orleans, LA.  (Click here for copy of paper.)

Reifman, A. (2001, October). Social influences on heavy-drinking college students’ readiness to change.  Society for the Study of Human Development. Ann Arbor, MI.

Reifman, A. (2001, August). Future orientation buffers the relationship between network and own drinking. American Psychological Association. San Francisco, CA.

McCourt, A., Watson, W., & Reifman, A. (2001, April). Gender differences in college drinking and its risk factors. All-University Conference on the Advancement of Women in Higher Education, Texas Tech University, Lubbock, TX.

Reifman, A., & Watson, W. (2001, February). Normative influence on college drinking: A longitudinal analysis.  Society for Personality and Social Psychology. San Antonio, TX.

Reifman, A., & Watson, W.K. (2000, August). Social networks and college drinking: A longitudinal study. American Psychological Association. Washington, DC.

We surveyed a random sample of entering Texas Tech University freshpersons beginning in the Fall 1999 term. Sampling was entirely from dormitory residents. Because the university requires first-year students to live in the residence halls (with some allowable exceptions), the sample can be considered largely representative of first-year students at the university as a whole. Surveying was done via campus mail (in both directions). A total of 3,283 freshpersons lived in the residence hall system at the start of the Fall 1999 semester. To accomplish our plan of sending out 600 questionnaire packets, we used the technique of systematic sampling with a random start (Babbie, 1998). Every fifth name was taken from an alphabetical list until enough cases were obtained (the starting point was in the A’s).

Written materials in the packet (cover letter, introduction to questionnaire, and consent form) alerted students that the project would involve three waves of measurement. They were offered payments of $10, $10, and $20 for completing each wave of the project (wave 3 has yet to begin). Students were informed that if they dropped out of the study at a wave, they could not re-enter. It was thought that giving the biggest compensation at the final wave would induce students to stay in the study. Also, based on findings from Willimack, Schuman, Pennell, and Lepkowski (1995) that a gift pen increased the response rate, we included a free Texas Tech pencil in all wave-1 packets.

Rogelberg and Luong (1998, Table 2) provided 10 tips for facilitating response to mail surveys. We only became aware of these tips after we began our study. Still, we managed to adhere at least in part to five of them (either in "letter" or in "spirit"):

• We made follow-up telephone calls to remind students to return their surveys.
• We provided incentives to complete the survey, as noted above. Rogelberg and Luong (Table 2) stated that is was "best to include the incentive with the survey." We included the pencil with the survey, but the monetary reward was given only after subjects’ returned the survey.
• Rogelberg and Luong urged the use of a stamped return envelope with the survey. The underlying idea would appear to be that subjects should not have to provide their own postage. This was accomplished through use of campus mail. For students who moved off-campus (which some had done at wave 2), we did provide a stamped return envelope.
• We made appeals in the cover letter in terms of the importance of subjects’ returning their surveys. We discussed the idea of representative sampling and how it was necessary to receive responses from all subjects to ensure the accuracy of the results.
• We indicated that the project was being sponsored by official "respected" institutions. The cover letter was written on Texas Tech University letterhead, the Principal Investigator’s (PI) faculty position was mentioned, and the sponsorship by a Federal government agency was mentioned in the consent form.

We did not adhere to three of Rogelberg and Luong’s guidelines:

• We did not notify individuals beforehand that they would be receiving a survey.
• We did not put "interesting" and "easy" questions first and demographic ones last. We put the demographic questions first, as opposed to the potentiallly more "threatening" alcohol-related items.
• We did not "[p]ersonalize the cover letter (with individualized salutation and handwritten signature)." Letters were addressed "Dear Student," and the PI signed only an original letter, which was then photocopied.

The remaining two tips seemed non-applicable: using "nonmetered postage (instead of bulk-rate mailing) for outgoing envelopes;" and using "stamped instead of business-reply envelopes for return of surveys."

Data were collected from 274 respondents at wave 1. There were 174 females (63.5 %) and 100 males. Because female respondents were over-represented in the present sample and males under-represented, relative to Texas Tech’s roughly 50/50 gender composition, and because the genders differed somewhat in their binge-drinking levels, the data were also analyzed with sample weights designed to equalize the proportion of males and females. Binge drinking was defined having at least one heavy-drinking episode (five drinks in a row for men, four for women) in the last two weeks (Wechsler, Davenport, Dowdall, Moeykens, & Castillo, 1994). With sample weighting to equalize representation of men and women in the present study, 45.6% of the overall sample were binge drinkers, compared to the figure of 44.7% that is obtained when no special weighting of males and females is done (i.e., the original proportions of males and females are left alone). Because of this great similarity, and for simplicity, all the remaining analyses in this report are done without sample weights. In terms of race/ethnicity, 235 (86%) wave-1 respondents were White, 22 (8%) Hispanic, 8 (3%) Black, 7 (2.6%)Asian/Pacific Islander, 1 Native American/Alaskan Native, and 1 other/unknown (each of the latter two comprised less than 1% of the sample).

The fact that the wave-1 response rate was below 50% is a cause for some concern. However, based on the aforementioned follow-up reminder calls and other information, we learned that there were at least 23 students who could not reasonably be expected to return the questionnaire: 10 students apparently dropped out of school during the study period (as evidenced by disconnected phone numbers and/or questionnaire packets returned unopened); 3 students were under 18 and thus ineligible to participate; 1 was a second-year student (the study was restricted to first-year students); and 9 students claimed not to have received a packet at a time when it was too close to the ending deadline to send them replacement packets (16 other students whose lack of a questionnaire was made known to us in time were sent replacement packets). Thus, we estimate our participation rate to be 274 / 577 = .47. Other college studies have obtained approximately 50% participation rates, such as the phone survey reported in Reifman and Dunkel-Schetter (1990), so the present return rate – especially with a mail survey – is not out of line.

Wave 2 took place during the Spring 2000 semester. Wave-2 packets could potentially be sent to all 274 wave-1 respondents. However, we learned in the spring that 22 students either were not currently registered (which we interpreted as dropping out, at least for the present time) or had not left a Lubbock address. We thus sent out 252 wave-2 packets, of which 191 were returned completed (76% retention out of the 252).

Wave 3 took place during the Fall 2000 semester.  Of the 191 wave-2 respondents, 34 appeared to have left the university by wave 3 (as indicated by questionnaire packets coming back unopened with either no known forwarding address or an address outside of the Lubbock, Texas area).  Thus, 157 potentially answerable survey packets were sent .  A total of 119 surveys were completed and returned. The wave-3 participation rate (out of the people who should have received the questionnaire packet) was thus calculated as 119 / 157 = .76.

Attrition appeared to be non-random. Males were systematically lost, as they constituted only 55 (28.8%) of the wave-2 sample, as opposed to 36.5% of the wave-1 sample. A gender (2: male, female) X retention status (3: retained in study, still in school but non-responding, left school/no address) chi-square test was significant, c ² (N = 274, 2 df) = 20.2, p<.001.

A clear trend was also present regarding differential attrition due to alcohol consumption and misuse. Using the measure of total ounces of absolute alcohol per day, individuals who we considered to have left school by wave 2 (not registered or no Lubbock address) drank more at wave 1 (M = .87) than did students who were still at Texas Tech but did not return a wave-2 survey (M = .63), with retained wave-2 participants being the lightest wave-1 drinkers (M = .49). Although a one-way Analysis of Variance (ANOVA) did not reach statistical significance, F (2, 267) = 2.08, p = .13, it clearly appears that attrition was linked to heavier drinking.

With the exception of the grid subjects used to list and describe their social networks, all responses were collected via a Scantron computer-scorable form. For each question, the subject could select an answer choice from 0 to 9, which he or she would "bubble in" with a No. 2 pencil. However, some of our questions allowed for answers greater than 9, such as number of times the respondent consumed five or more drinks in a row in the last two weeks, which could range from 0 to 14. In this case, we directed subjects to use two Scantron items to answer one question. For the aforementioned item, the instructions read: "Use scantron items 23 and 24. If, for example, you had five or more drinks in a row 04 times, you would enter a ‘0’ for item 23 and a ‘4’ for item 24." An additional sheet that visually illustrated the use of two Scantron items to respond to a single question was also included in subjects’ packets at wave 1.

The vast majority of respondents appeared to understand this procedure, as evidenced by the small number of out-of-range responses. Other indicia of misunderstanding were rare, such as bubbling in only one of the two lines for a question (a subject who did not engage at all in a behavior still should have bubbled in a zero on each of the two lines). Based primarily on out-of-range values and visual indicia of misunderstanding, data for a small number of participants were deleted for some items. For example, the wave-1 variable of total ounces of absolute alcohol consumed per day (which was based on quantity and frequency items for beer, wine, and liquor) had 4 missing cases. Sample size (N) values for specific variables can be obtained in our various reports (convention papers and articles).

Another type of apparent error occurred when subjects gave a non-zero frequency of drinking a beverage in the last two weeks but a zero quantity for how much they drank, or a non-zero quantity but zero frequency. For example, a cross-tabulation of the wave-1 beer quantity and beer frequency variables revealed that an additional 13 respondents exhibited this type of discrepancy. Because deletion of these subjects’ data on these variables would have led to an even greater reduction in sample size for many analyses, data with the "zero/non-zero error" were kept in the study. Also, because the formula for calculating total ounces of absolute alcohol consumed per day involves multiplying quantity and frequency (or variables derived from them) together for each beverage, the presence of a zero will guarantee that beverage-specific consumption will be zero. This will likely lead to a more conservative (i.e., lower) estimate of alcohol consumption for subjects with the zero/non-zero discrepancy. This should work against our finding significant relationships between drinking and other variables as, if subjects truly consumed more than the value they received in the data set, this will result in measurement error.

The data were examined for outliers on the major drinking variables. In looking, for example, at a scatter plot of total ounces of absolute alcohol consumed per day at waves 1 and 2, one respondent stood out apart from all the others in terms of high consumption at both waves. The effect of this outlier was examined by computing a Pearson correlation between wave-1 and wave-2 consumption, both with and without the outlying case. Presence of the outlier increased the correlation (r = .79) compared to when the outlier was removed (r = .70). As noted above, our general approach has been to take actions regarding data management that will be most conservative (i.e., make it less likely to find significant relationships). Even though the outlying case inflates the cross-temporal stability correlation of total alcohol consumption, however, it was kept in the data set because it makes other, more substantive analyses more conservative. Many of the prominent analyses in this project involve prospective prediction of drinking outcomes by social/psychological variables (e.g., drinking level of one’s social network). For example, a wave-1 social variable would be used to predict a wave-2 drinking variable, controlling for the wave-1 version of the drinking variable. By inflating the relationship between wave-1 and wave-2 drinking, therefore, we are making it harder for the wave-1 social variable to predict the wave-2 drinking variable.

An excellent discussion of how to treat outliers (from Professor Peter Westfall) is available at:  http://www2.tltc.ttu.edu/westfall/images/5349/outliers_what_to_do.htm

As illustrated above, our tendency when observing "deviant" data points (out-of-range values; logical discrepancies between corresponding items [e.g., frequency with no quantity]; outliers) has been to either leave the data alone or to delete cases on a given item. We have not attempted to come up with alternative values, such as capping an outlying value at a lower, though still high, value.

Many of our analyses involve structural equation modeling with the AMOS program (AMOS 4; Arbuckle, 1994-1999). AMOS’s default procedure for missing data, involving the estimation of means and intercepts, was used to allow the maximum N for a given wave to be used (274 at wave 1, 191 at wave 2). Our main data file contains 274 cases; data lines for subjects who did not participate at wave 2 are full of blank spaces on the wave-2 variables. In an early attempt to run a two-wave model from this file, we discovered that AMOS filled in all the missing data to give us a two-wave N of 274. In order to restrict the sample for a two-wave model to only the 191 subjects who actually responded to both waves, we simply created an additional data file containing lines for only the two-wave participants. As a result, 191 will be the N on two-wave studies, and only small numbers of missing-data values will be filled in by the program (e.g., for subjects who left a particular item blank, or who had a response excluded for being out of range).

Because research has shown college-student drinking to vary by time of the semester (Del Boca, Greenbaum, Darkes, & Goldman, 2000), we evaluated this issue in our data set. From the outset, we have recorded the arrival date of each returned survey packet, which we then converted into a variable (i.e., day 1, 2, 3, etc., of a given wave’s study period). Complicating matters is the fact that even though we delivered all packets for distribution at the same time, different residence halls received them at different times (delays were on the order of a few days to a few weeks).

Given the above considerations, we conducted a series of Analyses of Covariance (ANCOVA) for wave 1, in which residence hall was an experimental factor, arrival date a covariate, and total alcohol consumption the dependent variable. Because most halls at the university are single-sex, the ANCOVA was done separately in males and females. Neither residence hall nor arrival date significantly predicted drinking in either gender. It should be noted, however, that for some cells the sample sizes were small (less than 10). Also, we obtained the correlation between arrival date and total alcohol consumption, separately by each residence hall. These correlations varied widely, from -.45 to +.46 and +.47. Nearly half (7/15) of the correlations were between -.17 and +.01. None was significant at p < .05 (two-tailed), given the small N values. On the whole, there thus appeared to be no systematic relationship between when a survey was returned and the amount of drinking reported.

In light of the above findings, neither residence hall nor arrival date was included as a control variable in any of the substantive analyses conducted in the project. Given these variables’ lack of significant relations to drinking, their omission would be unlikely to cause specification error.

A Certificate of Confidentiality was obtained from the U.S. Department of Health and Human Services. The Certificate protects the investigators from being forced to release any research data in which subjects are identified, although there are some limited exceptions. The Certificate was described to subjects as part of the consent form, including examples of areas in which the protection against disclosure of their data would not be absolute.

Arbuckle, J.L. (1994-1999). Amos, Version 4. Chicago, IL: SmallWaters.

Babbie, E. (1998). The practice of social research (8^th Ed.). Belmont, CA: Wadsworth.

Del Boca, F.K., Greenbaum, P.E., Darkes, J., & Goldman, M.S. (2000, August). College student drinking growth curves: Linear trends and environmental peaks. Poster presented at the 108^th Annual Convention of the American Psychological Association, Washington, DC.

Reifman, A., & Dunkel-Schetter, C. (1990). Stress, structural social support, and well-being in university students. Journal of American College Health, 38, 271-277.

Rogelberg, S.G., & Luong, A. (1998). Nonresponse to mailed surveys: A review and guide. Current Directions in Psychological Science, 7, 60-65.

Wechsler, H., Davenport, A., Dowdall, G., Moeykens, B., & Castillo, S. (1994). Health and behavioral consequences of binge drinking in college: A national survey of students at 140 campuses. Journal of the American Medical Association (JAMA), 272, 1672-1677.

Willimack, D.K., Schuman, H., Pennell, B., & Lepkowski, J.M. (1995). Effects of a prepaid nonmonetary incentive on response rates and response quality in a face-to-face survey. Public Opinion Quarterly, 59, 78-92.

LINKS TO OTHER ALCOHOL- AND SUBSTANCE-RELATED SITES ON THE WEB
(WITH AN EMPHASIS ON ADOLESCENT AND COLLEGE-STUDENT DRINKING)

National Institute on Alcohol Abuse and Alcoholism

NIAAA College Drinking Website

National Institute on Drug Abuse

Substance Abuse & Mental Health Services Administration

Texas Tech University's Center for the Study of Addiction & Recovery (not associated with my research, but a valuable campus resource)

Monitoring the Future Study -- National surveys of high school and young adult substance use conducted by University of Michigan

College Alcohol Study --  National surveys conducted by Harvard University School of Public Health

My lecture notes on "Alcohol and Other Substance Use" for my courses...

• Problems of Adolescence (HDFS 3317)
• Development in Young Adulthood (HDFS 3318)

Research Institute on Addictions -- Buffalo, New York -- My old home (1991-1997)

Mothers Against Drunk Driving (MADD)

DUI Foundation

American Council for Drug Education

Indiana Prevention Resource Center

The Blood Alcohol Educator

Center On Addiction and Substance Abuse

Campaign for Tobacco-Free Kids

Partnership for a Drug-Free America

Higher Education Center for Alcohol and Other Drug Prevention

National Social Norms Institute (attempts to reduce heavy drinking and other risky behaviors by challenging norms that "everyone is doing it")

Excellent introduction to social networks from the International Network for Social Network Analysis (INSNA)