Copyright © 2011, New Forums Press, Inc., P.O. Box 876, Stillwater, OK 74076. All Rights Reserved. The Transfer Velocity Project: A Comprehensive Look at the Transfer Function Craig Hayward Planning, Research, and Knowledge Systems, Cabrillo College The 1999-2000 Transfer Velocity Project (TVP) cohort of 147,207 community college students is used to develop both a college-level endogenous model, appropriate for applied research and guidance for campus action, and a student-level model. Survival analysis (Cox regression) is employed to evaluate the relative contribution of 53 student-level characteristics to the transfer outcome. Results suggest practical interventions for enhancing the transfer function such as encouraging the completion of the associate degree en route to transfer as well as enhancing institutional transfer culture by, for example, fully staffing campus-based transfer centers. Discussion focuses on the identification of transfer-facilitating characteristics, the strengths and weaknesses of the current research, suggestions for future research, and practical approaches to facilitating transfer. Introduction Transfer rates are the bread and butter of transfer studies, state accountability systems, and individual institutional research offices. They are accepted and familiar, though they may not be as informative as commonly believed. Transfer rates function well as key performance indicators that shine a spotlight on trends over time, but they are susceptible to misinterpretation due to the incomplete picture they provide. One problem with transfer rates is that they ignore any transfer activity past the time window used to define the rate – potentially a great deal of transfer activity. Recently published research tracking high-risk cohorts that began postsecondary education in the 1970s has shown that transfer and postsecondary degree completion is actively occurring over multiple decades (Attewell, Lavin, Domina & Levy, 2007). Researchers tracked nearly 2,000 non-traditional, low-income women through the integrated system of community colleges and universities in the City University of New York (CUNY) system. These students entered the system between 1970 and 1972 at a time when the CUNY had just switched to an open admissions policy. Over a 30-year time frame more than 70% of the women had graduated from CUNY. However, about a quarter took more than 15 years to finish. What, then, is the best time frame to use for calculating a transfer rate? Consider for a moment the variety of transfer rates that are in use. The official Integrated Postsecondary Educational Data System (IPEDS) rates allow new community college students up to three years to transfer, though that time frame will soon move to four years – a nominal 200% of the “normal time to completion” for two-year college students (IPEDS, 2010). Researchers, state agencies, and policy analysts who make community colleges their business often use five- or six- year transfer rates (CCCCO, 2008; CPEC, 2007; Sengupta & Jepsen, 2006; Shulock & Moore, 2007). Recently, some policy analysts have begun to evaluate student outcomes using longer time frames such as seven years (Offenstein, Moore, & Shulock, 2010). However, the rationale for using a given time frame is seldom explicitly evidence-based. While transfer rates with longer time frames are better indicators of the true rate for a given cohort, reliance on a single time frame and/or enrollment at a single institution still foreshortens our perspective and our understanding of transfer. College transfer rates or even system rates cannot provide a complete picture of the transfer function. Shifting focus to the individual student provides a different perspective; one that can be even more useful in terms of yielding results that can be applied by practitioners. In The Toolbox Revisited, Adelman (2006) followed a cohort of students across all of their postsecondary enrollments, allowing up to 8.5 years for students to complete their postsecondary education. He found that nearly 60% of his those in NELS:88/2000 cohort who attended college, attended more than one postsecondary institution, calling into question the validity of focusing on any one institution’s rates without reference to the greater tapestry of student postsecondary engagement. The current study provides both an institutional and a student perspective. The student-level model of transfer Vol. 18, No. 2, Spring 2011 / 21 utilizes survival analysis to evaluate how student coursework and behavior enhances or detracts from students’ transfer velocity. Combining the complementary, but somewhat different perspectives of the college-level focus on transfer rates and the student level focus on velocity provides a richer, subtler and more fully dimensional understanding of how our educational systems and our diverse student body interact and influence transfer rates. In referring to transfer velocity, we borrow the term “velocity” from physics. In the physical world an object’s velocity is determined by its speed and orientation as it travels along a pathway (or vector). Although analyses of both transfer rates and transfer velocity require the tracking of cohorts of similar students, the velocity analogy provides additional insight beyond the traditional focus on a specific time window (e.g., the six-year transfer rate). This is because transfer occurs as a more or less continuous function over time. Transfer velocity, then, is a way of understanding how the characteristics of a postsecondary system interact with the characteristics and behaviors of students to enhance or impede student progression over time along a pathway toward a goal. The key advantage of transfer velocity over transfer rates is that more information – and more actionable information – is used and represented in the analysis of students’ transfer velocity. Methods When assessing transfer rates and transfer velocity, determining which students to track and study is a key issue. Each California Community College has a multipronged mission statement, reminding us that not all students are transfer-oriented. As such, there is general agreement in recent research that, students who are not transfer-oriented, because they hold other goals and are on other trajectories, should be excluded from the analysis, if possible. (Adelman, 2005; Banks, 1990; Cohen, 2005; CPEC, 2007; Driscoll, 2007; Horn & Lew, 2007a; Bahr, Hom & Perry, 2005; Shulock & Moore, 2007; Sengupta & Jepsen, 2006; Spicer & Armstrong, 1996; Thompson, 2002). Community colleges typically host multiple functions. For instance, students may be seeking to acquire job skills and enter directly into the workforce either with or without a certificate of achievement or an associate’s degree. As Thompson (2002) suggested, the cohort that is tracked in any given study should be selected based on the goal of the research. To this end, the TVP uses the “behavioral intent” method established by Bahr, Hom & Perry (2005). How did students qualify to be included in a TVP cohort? The “behavioral intent” method involves tracking cohorts of first-time college students for six years to determine if they show behavioral intent to transfer, also referred to as “serious academic intent” (Bahr, Hom & 22 / Journal of Applied Research in the Community College Perry, 2005; Woo, 2009). A student becomes potentially eligible to enter a cohort by enrolling for the first time at any California Community College (CCC). Students are assigned a cohort year according to the academic year in which they first enroll at a CCC. Students who have completed at least 12 credit units and who have enrolled in (not necessarily passed) a transfer-level math and/or a transfer-level English course within six years of initial enrollment enter into the TVP cohort for that year. The analyses in this article focus on the cohort of students that first entered the CCC system in the 1999-2000 academic year. The 1999-2000 TVP cohort was selected for model building because it provides sufficient time to track the vast majority of transfer activity; of all the transfer activity that occurred within 15 years of initial entry, 95% had occurred within 9 years of initial enrollment. Additionally, major sources of data for the model building (e.g., Census 2000) were gathered in the time period when 1999-2000 TVP cohort students were most active. For purposes of calculating institutional transfer rates, students were assigned to a home college according to where they first enrolled. In cases where students simultaneously enrolled at more than one community college in their first postsecondary semester (3% of cases), the tie was broken by random assignment to a single home college. Nine-year transfer rates were used for the college-level model because they capture the vast majority of TVP cohort transfer activity. Cohort Description There are 147,207 students included in the 1999-2000 TVP cohort, representing 24% of all first-time freshmen for that academic year. The students in the 1999-2000 TVP cohort had a mean age of 20 (median age of 18) upon initial enrollment in the CCC. Forty-four percent of the cohort is male, 25% had initially enrolled at a CCC while still in high school, 40% had attended more than one CCC, 40% had received some form of financial aid, 52% had transferred within nine years of initial entry, 29% earned an associate’s degree within nine years of initial entry, and 6% had earned a certificate of achievement. The ethnic breakdown of the 1999-2000 TVP cohort is 42% white, 24% Latino, 14% Asian, 6% African American, 4% Filipino, 1% Native American, 3% other, and 7% unreported. Data Sources The TVP utilized three major data sources. The first was the CCC Chancellor’s Office (CCCCO) Management Information Systems (MIS) data base, a central data repository for all CCCs which includes all student, college and course information that has been submitted annually since 1992. The second data set comprised 45 sociodemographic variables that describe the social and envi- ronmental context of each community college in California. It was derived largely from Census 2000 data indexed to individual college’s service areas and was also provided by the CCCCO (van Ommeren, Liddicoat, & Hom, 2008). The third data source was the transfer center surveys that are required to be submitted to the Chancellor’s Office annually by the transfer center director (or proxy) at each CCC. The annual transfer center survey contains a variety of information from the allocated budget and funding sources, to the staffing levels, number of student contacts, and ratings of areas of significant challenge. Data from three years were aggregated by TVP staff to form a data set describing how transfer center practices, staffing, student contacts, priorities and more varied around the state for the time period under examination (i.e., 1999-2008). Results An Endogenous Model of College Transfer Rates The endogenous college-level model focuses on those factors that are likely to be under the control of community college personnel. This perspective on college-level factors diverges from an accountability-based perspective in that the ability of college staff to control or manipulate the factors in the model is more important than the overall explanatory power or fit of the model. We seek to understand those factors or levers that are close at hand and that might be adjusted through a college’s policies and actions to increase the transfer success of students The Transfer Center survey data and Chancellor’s Office (CO) MIS data about college schedules and multiple community college attendance patterns were combined in a backward linear regression analysis to determine which factors best predicted nine-year college-level transfer rates (all variables were included initially and variables that did not provide a unique contribution to the model’s explanatory power were removed in blocks). In the bestfitting endogenous model (adjusted R2 = 0.48), six factors accounted for 48% of the variability in college-level transfer Transfer Velocity Project rates. While the R2 value is smaller than that of the exog- enous accountability model created by the CO (CCCCO, 2008), the endogenous model can still be considered to be robust and useful. Importantly, the factors in this model provide guidance and can be used to galvanize a call to action by campus personnel. In Table 1, the standardized coefficients (i.e., Beta values) show the direction of association as well as allowing for a comparison of the relative strength of the variables included in the regression model. The Transfer Center surveys provide three significant endogenous predictors of college-level transfer rate: 1) the number of CSUs with which the college has established a TAA pathway; 2) the average annual number of students who sign a TAA with a UC; and 3) the Full Time Equivalency (FTE) of the college’s Transfer Center Director. The signs of all three of these factors’ Beta coefficients are positive, meaning that as the number of TAAs and the FTE of the Transfer Center director increase, so does a college’s transfer rate. The TAA agreements, particularly the numbers of CSU partners, have a strong relationship with transfer rates, as the Beta coefficient of .212 indicates. The college scheduling variables show that, perhaps unsurprisingly, the percentage of transferable course sections at a college is positively associated with college’s nine-year transfer rate. Nearly as strong, but in the opposite direction, is the relationship between the percentage of Career Technical Education (CTE) sections offered by a college and the college transfer rate. The final variable in the model is the average level of college swirl, as indicated by the average number of colleges attended by the student body; this variable is positively associated with transfer indicating that colleges with higher proportions of students attending multiple community colleges tend to have higher transfer rates. Student-Level Model of Transfer Velocity Fifty-three student-level characteristics were analyzed with a form of survival analysis known as Cox regression. Cox regression is designed to properly handle data where the ultimate disposition of all cases is not known in regards to a key outcome. Originally developed to analyze medical research, a Cox regression analysis outTable Endrogenous college-level transfer model coefficients Table 1. 1. Endogenous college-level transfer model coefficients puts risk statistics, which show Unstandardized Standardized a the relative degree of risk Model Coefficients Coefficients t Sig. (compared to an average coBeta B Std. Error hort member) that is associ(Constant) .144 .080 1.940 .075 ated with the presentation of Pct. Transferrable Sections .241 .219 .077 2.868 .000 a particular characteristic. A Pct .CTE Sections -.225 -.222 .086 -2.579 .005 Student TAAs signed (UC) heightened “risk” of transfer .201 .001 .000 2.551 .011 No. of CSU with TAAs indicates that a certain charac.212 .041 .014 2.842 .012 FTE of Center Director teristic or program is positively .154 .055 .027 2.009 .047 Average swirl level .279 .099 .026 3.766 .001 associated with transfer. a. Dependent Variable: Nine-year College Transfer Rate In the TVP the risk of the Vol. 18, No. 2, Spring 2011 / 23 terminal outcome is actually a positive event – the transfer 2 reveals that there is a small negative effect of -5.6% asof a CCC student to a baccalaureate-granting institution. sociated with being male. That is, males are 5.6% less likely Cases that do not demonstrate the terminal outcome of to successfully transfer than females, after controlling for interest (i.e., transfer) by the end of the observation pe- the effects of the other variables in the model. In addition to the simple binary variables, there are riod are considered censored because transfer could still Table 2. Student-level transfer velocity model occur, but it would occur outside our window of observa- three important student characteristics that are presented tion. Cox regression is designed to properly model and as sets of binary variables. The three sets of variables are: account for censored cases. Table 2. Student-level transfer velocity model Properly modeling censored Incidence/ B Standard Sig. Exp(B) Relative risk of observations and providing average Error (B) transfer discrete information about Age 20.1 -0.0379 0.0008 0.000 0.963 -3.7% a large number of student Male 44.3% -0.0579 0.0076 0.000 0.944 -5.6% characteristics are strengths of the Cox Regression proceAsian (reference group) 13.7% dure. However, Cox RegresAfrican American 5.7% -0.1662 0.0190 0.000 0.847 -15.3% sion may not be as familiar as Ordinary Least Squares Filipino 3.7% -0.1590 0.0212 0.000 0.853 -14.7% regression to many readers Latino 24.0% -0.3083 0.0127 0.000 0.735 -26.5% and it may require some Native American 0.9% -0.4368 0.0472 0.000 0.646 -35.4% initial effort to interpret Other 2.2% -0.0762 0.0251 0.002 0.927 -7.3% the results. Fortunately, the Pacific Islander 0.7% -0.1778 0.0455 0.000 0.837 -16.3% Cox regression procedure White 42.0% -0.2020 0.0112 0.000 0.817 -18.3% is capable of producing a relatively intuitive statistic, Unknown 7.0% -0.2163 0.0165 0.000 0.805 -19.5% namely the relative risk statistic which shows whether a Transfer only 27.8% 0.3326 0.0083 0.000 1.395 39.5% given student characteristic Intellectual growth 10.7% 0.2589 0.0110 0.000 1.296 29.6% enhances, reduces or has no Basic Skills 4.0% 0.1503 0.0183 0.000 1.162 16.2% effect on the odds of successGED 9.1% 0.0529 0.0131 0.000 1.054 5.4% ful vertical transfer. With the exception of Goal Unknown 1.2% 0.0332 0.0352 0.345 1.034 3.4% age and GPA, the variables Job Skills goal 15.8% 0.0130 0.0110 0.239 1.013 1.3% in the student-level model Undecided goal 38.6% -0.0001 0.0082 0.992 1.000 0.0% are expressed as binary opDegree and transfer 59.0% -0.0158 0.0082 0.052 0.984 -1.6% posites where 1 = “CharacDiscovery goal 7.7% -0.0286 0.0142 0.044 0.972 -2.8% teristic is present” and 0 = Certificate only 4.0% -0.2118 0.0221 0.000 0.809 -19.1% “Characteristic is absent.” The average of a binary Maintain license 2.9% -0.2223 0.0255 0.000 0.801 -19.9% variable across all students Degree only 13.8% -0.4419 0.0136 0.000 0.643 -35.7% is equal to the proportion of cases with a value of one. GPA 2.92 0.3422 0.0072 0.000 1.408 40.8% For example, the average of <20% of grades are "W" 46.0% 0.4633 0.0084 0.000 1.589 58.9% the binary variable “Male” = .443, therefore 44.3% of the 1999-2000 TVP cohort <10% of units are CTE 49.6% 0.1296 0.0077 0.000 1.138 13.8% is male. The relative risk of Special Admit history 24.5% 0.1071 0.0108 0.000 1.113 11.3% simple binary variables like Summer enrollment 69.3% 0.3859 0.0090 0.000 1.471 47.1% gender is straightforward; Full Time student 21.8% 0.2732 0.0097 0.000 1.314 31.4% when the characteristic is Attended > 1 CCC 40.3% 0.2063 0.0085 0.000 1.229 22.9% present (i.e., the student is a male) the risk in the appropriate column of Table First English, Transfer lvl. 50.4% (reference group) 2 applies. Consulting the First English, Degree lvl. 22.8% -0.1103 0.0101 0.000 0.896 -10.4% relative risk column in Table 24 / Journal of Applied Research in the Community College 1) Ethnicity; 2) First English course; and 3) First math course. The interpretation of relative risk for the binary variables in these sets needs to be in the context of the other members of the set. This is because in order for the model to produce meaningful output for these variable sets, it is necessary to omit one member of the set; the omitted variable serves as a reference category when interpreting the remaining variables. For example, Asians were omitted from the ethnicity set; as the ethnic group with the highest average transfer rate, Asian students constitute a good reference group. The relative risks of the other ethnic Table 2 (continued) B First English, Basic skills Incidence/ average 15.7% -0.1379 Standard Error (B) 0.0127 First English, other 1.9% -0.0704 0.0267 No English at CCC 9.3% -0.1018 0.0145 First math, Transfer lvl. (reference group) First math, Degree lvl. 25.4% 36.3% -0.3908 0.0097 First math, Basic skills 22.6% -0.6406 0.0128 Math, other 2.8% -0.4933 0.0235 No math at CCC 12.8% -0.8198 0.0151 AA early 9.1% 0.7061 0.0122 AA mid 14.3% 0.3980 0.0101 AA late 5.7% 0.0242 0.0167 Certificate 6.4% -0.5169 0.0184 Transfer Engl. in Year 1 18.7% 0.1692 0.0100 Transfer math in Year 1 6.7% 0.1583 0.0138 Postponed Engl. & math 15.2% -0.3985 0.0119 Loan* 3.7% 0.1316 0.0211 Work Study* 3.2% 0.1067 0.0224 Grants (e.g., CalGrant) 16.2% 0.0648 0.0167 Scholarship 2.6% 0.0576 0.0220 Pell Grant 25.6% -0.0228 0.0157 BOG 39.2% -0.0514 0.0115 Skills course 1.8% -0.0022 0.0279 Guidance course 15.6% -0.0619 0.0104 Reading course 15.7% -0.1733 0.0119 EOPS 11.9% -0.0804 0.0158 DSPS 2.8% -0.0380 0.0220 CalWorks 3.9% -0.0894 0.0261 groups are expressed in relation to the Asian group and are all negative (because members of other ethnic groups have lower odds of transfer than Asian students). Whites, for instance, are 21% less likely to transfer than Asians, all other things in the model being equal. The omitted category for first course in math and first course in English is the transfer level course. Thus, it is possible to know that students whose first math class is a basic skills course are 49% less likely to transfer than students whose first math class is transfer math, after controlling for the effects of the other variables in the model. In addition to the binary variables, there are two variables that are expressed Sig. Exp(B) Relative risk of as a range of values: age and transfer GPA. These variables are in0.000 0.871 -12.9% terpreted in a similar fashion 0.008 0.932 -6.8% to the binary variables: for 0.000 0.903 -9.7% each increment in age or GPA the relative risk increases by the given percentage. Therefore a student with an “A” 0.000 0.676 -32.4% average (GPA = 4.0) is 42.3% 0.000 0.527 -47.3% more likely to transfer than a 0.000 0.611 -38.9% student with a 3.0 GPA and 0.000 0.441 -55.9% 85% more likely to transfer than a student with “C” average or 2.0 GPA (42.3% x 0.000 2.026 102.6% [4.0-2.0] = 84.6%). Similarly 0.000 1.489 48.9% a student who enters the co0.148 1.025 2.5% hort at age 18 is 38% more 0.000 0.596 -40.4% likely to transfer than a TVP student who first enrolled at community college at age 28 0.000 1.184 18.4% (3.8% x [28-18] = 40%). 0.000 1.172 17.2% The output of the stu0.000 0.671 -32.9% dent-level transfer velocity model is presented in Table 0.000 1.141 14.1% 2. The first column shows the 0.000 1.113 11.3% student-level characteristic in question while the second 0.000 1.067 6.7% column shows the percent0.009 1.059 5.9% age of the 1999-2000 TVP 0.146 0.977 -2.3% possessing that characteristic. 0.000 0.950 -5.0% In the cases of GPA and age, the second column shows 0.936 0.998 -0.2% the mean GPA or mean age upon cohort entry. The third 0.000 0.940 -6.0% column shows the Beta value 0.000 0.841 -15.9% (“B”) of the characteristic; a positive value indicates that 0.000 0.923 -7.7% the characteristic is positively 0.085 0.963 -3.7% associated with successful 0.001 0.915 -8.5% transfer while a negative Vol. 18, No. 2, Spring 2011 / 25 value means the characteristic is negatively associated with transfer. The next column shows the standard error (“SE”) of the Beta value. The standard error is used in calculating the statistical significance. The next column gives the statistical significance of the effect associated with the characteristic. When the significance value is smaller than 0.01 (i.e., p < .01) the associated effect is considered to be statistically significant or not likely to have occurred by chance. The column after the significance applies the exponential function, Exp(B), to the Beta value to derive the relative risk. The final column simply makes the relative risk easier to interpret by subtracting one from Exp(B) and then expressing the difference as a percentage. This last column is the most important column for interpreting the effects of the variables in the student-level transfer model. How large does a relative risk have to be in order to be meaningful? The size of the 1999-00 Transfer Cohort (n =147,207) means that statistical significance is a low bar for evaluating the importance of an effect. Indeed, the lack of significance may itself be of interest in some cases, such as in the case of CalWORKs and DSPS; the lack of significance at p < .01 indicates that, all other things being equal, the transfer rates of students in these programs are statistically indistinguishable from the transfer rates of students who are not in the program; this could be considered to be evidence of a successful intervention as students in those programs are typically overcoming extra barriers on their path to transfer. Determining whether an association is large enough to be meaningful is essentially a subjective judgment dependent upon a number of contextual factors. While others may prefer a different threshold, we highlight in bold those factors that have a relative risk of at least +/-10%. Limitations of the Study While the use of population data makes for a high level of confidence in our findings, there are certain limitations inherent in the current approach. First, despite the extensive set of information available for multivariate analysis, there is a notable gap in the area of socioeconomic information. While income and socioeconomic information is available for those students who apply for financial, only about half of the 1999-2000 TVP cohort completed a financial aid application. The lack of socioeconomic information results in the confounding of socioeconomic status with financial aid and ethnicity, making results for these areas difficult to interpret. A second limitation of the current analysis is that, despite the strongly multivariate nature of the analyses, these are essentially correlational analyses. Inferences about causation should be made cautiously and with reference to other research and information sources. Finally, because we are interested in understanding transfer behavior among those who clearly demonstrate an 26 / Journal of Applied Research in the Community College intention to transfer, the current approach trades off some generalizability for a more incisive look at a the population of interest. The fact that the TVPs transfer models replicate a wide array of prior research gives us confidence in their general soundness and also in the soundness of the TVPs more novel findings. Discussion and Conclusion The distinct elements found in the quantitative analysis of the current study correspond to the factors identified in the qualitative analysis of research on colleges with higher than expected transfer rates. (Schiorring & Mery, 2008). The six transfer-promoting factors identified by Schiorring and Mery (2008) are: 1) Transfer culture; 2) Student-focused environment; 3) Commitment to the institution; 4) Strong, strategic high school relationships; 5) Strong four-year college relationships; and 6) Effective support services Three of the six factors in particular are corroborated by the endogenous model: the importance of a “transfer culture” (as indicated by a high proportion of transfer sections in the class schedule); the importance of “strong four-year college relationships” (as indicated by the number of TAA pathways); and “effective support services” (as indicated by the FTE of the Transfer Center Director and the number of signed TAAs). The data that were available for use in the endogenous model did not sufficiently represent the other three factors, so the current endogenous model is not able to speak to their relative importance. These untested factors may be equally important, more important or less important, than the factors that are included in the current model. The endogenous model suggests that effective student support services (in this case, an adequately staffed Transfer Center), strong relationships with four-year schools (large number of TAAs), and a course schedule that has a high proportion of transferrable coursework (relative to CTE courses and non-transferable courses) are transfer-promoting factors that are within the control of campus personnel. A student body with a high proportion of students who attend multiple community colleges is also positively associated with college-level transfer rates. This factor is picked up in the student-level model, as well, and is discussed in more detail in that section. Earlier research has also highlighted the importance of some of the factors in the endogenous model. Poisel and Stinard (2005) examined the importance of interinstitutional networks for transfer success. Of particular note was the importance of the relationships between faculty and staff at the university and those at the community college. Adelman (2005) found that students who attended multiple community colleges were more likely to transfer. A sustained, strategic focus on the six areas highlighted by the endogenous model could result in higher transfer rates over time. It is worth noting that Schiorring and Mery (2008) state that transfer culture extends beyond the college Transfer Center; we agree with this point and do not feel that an exclusive focus on Transfer Center characteristics will provide the best understanding of inter-collegiate differences in transfer rates. However, while the Transfer Center may not be the totality of college’s transfer culture, it is likely to be a meaningful barometer of it. Discussion of the student-level transfer velocity model, presented in Table 2, is broken into the following areas: Demographic characteristics; Educational goals; Academic performance; Enrollment patterns; Coursetaking patterns; Degrees and certificates; Financial aid; and Special programs. Asians have the greatest odds of transferring as a group and thus are the reference group used to interpret the relative risk of ethnic group membership. Native Americans and Latinos have the greatest negative relative risk statistics (-35.4% and -26.5%, respectively). These are among the largest negative relative risk statistics in the model. This outcome is particularly important as Latinos are the second most populous ethnic group in the transfer cohort and in the CCC system as a whole. Raising the transfer success of Latinos and Native Americans should be aspects of any plan to improve Baccalaureate attainment in California. Educational Goals tra ns is fe co r ve ry C G er oa tif l ic at e M on ai nt ly ai n li c en se D eg re e on ly G oa l D an d eg re e n oa l G id ed nd ec D U Jo b Sk i ll s nk no w G ED G oa lU ct u el le In t Tr a ns f er o nl y al gr ow th B as ic Sk il l s Student intention and motivation is an important part of student-level model of transfer. Students’ goals may change over time and thus a student may have Demographic Characteristics multiple goals over their postsecondary career. The At the top of Table 2, one can see that increased age educational goal variables in the transfer velocity model is negatively associated with transfer. While the effect for therefore indicate whether a student has ever expressed a one-year difference is not large, students who are 10, 20, a given goal. For example, 59.0% of the 1999-2000 TVP or 30 years older than is typical are much less likely to cohort students declared a goal of obtaining an associate’s transfer (-37%,- 74% and -111%, respectively). This finding degree and also transferring to a four-year school (i.e., was present in the exogenous college-level model as well, “degree and transfer”) at least once in their academic cawhere colleges with older student bodies tended to have reer at the CCC, but some students who declared “degree reliablyVelocity lower transfer Transfer Project rates (model not shown). Gender and transfer” may have also declared another goal such as was also found to have a small negative effect on the odds improving their basic skills (i.e., “basic skills”) in which Figure 1. Relative risk associated with ever declaring given educational goals of transfer (-5.6%). case the relative risk of both goals would apply to them. Relative risks are cumulative such that the relative risk of a student 50% who had declared a goal of “basic 39% 40% skills” and “degree and transfer” would be 14.6% (16.2% - 1.6% = 30% 30% 14.6%). Table 2 and figure 1 show that 16% 20% there are three educational goals, which are relatively potent predic10% 5% 3% 1% tors of successful transfer: Transfer 0% 0% only (39%); Intellectual growth -2% -3% (30%); and Basic skills (16%). Hav-10% ing an educational goal that purely focuses on transfer is, reasonably -20% -19% -20% enough, positively associated with -30% successful transfer. Perhaps somewhat more surprising is the finding -36% -40% that a goal of intellectual and cultural development is also positively associated with transfer. One might think that a goal of intellectual and cultural development would be a characteristic of lifelong learners Figure 1. Relative risk associate with ever declaring given educational or students who already possess a postsecondary degree and are regoals. turning to the CCC for intellectual Vol. 18, No. 2, Spring 2011 / 27 stimulation. Here it is important to remember that we are analyzing a special cohort of students. Students who already possess a postsecondary degree are excluded from the transfer cohort by definition, and are therefore not included in the model. When interpreting this finding we should be cognizant of this exclusion by phrasing the finding thus: the students who demonstrate that they intend to transfer and who declare the educational goal of “intellectual and cultural development” are 30% more likely to transfer than those TVP students who do not declare that goal. The 16% boost in the odds of transfer that is associated with a student indicating a goal of basic skills is intriguing. It suggests that a focus on improving basic skills is congruent with successful transfer, at least for those students who are able to progress far enough to complete 12 credit units and attempt a transfer level English and/ or math class. The majority of students in the 1999-2000 TVP cohort (59%) declared “Degree and transfer” as a goal at some point. Therefore, the goal of “Degree and transfer” is not strongly associated with successful transfer because it does not distinguish among students. Those goals which have strong relative risks associated with them are less common; the six educational goals that have the largest relative risk statistics are indicated by only about a quarter of the cohort or less. The three educational goals, which are negatively associated with transfer are: Certificate only (-19.1%); Maintain license (-19.9%); and Degree only (-35.7%). These negative associations may indicate that there are some students inappropriately included in the TVP cohort. That is, the TVP cohort is behaviorally defined in such a way as to focus on students who are likely pursuing transfer. However, the cohort definition is not perfect. Students with these three goals are less likely to transfer than other TVP students and some of them, particularly the 13.8% of the TVP cohort who indicate that they only want an associate’s degree, may be inappropriately included in the TVP. Those students who are expressly seeking terminal associate’s degrees are distinct from those who indicate a goal of degree attainment and transfer. The negative association between obtaining a certificate achievement and transfer supports this interpretation, as well. Academic Performance A student’s competencies and course-level success has a definite effect on successful transfer. An increase of one grade point in GPA is associated with a 41% greater chance of transfer. Thus, a straight “A” student is 41% more likely to transfer than a student with a “B” average. Other grades are important, as well. Of particular relevance is the “W” grade which is given when a student withdraws in the middle of a class. Around half of the TVP 28 / Journal of Applied Research in the Community College students (46%) had transcripts in which 20% or more of their enrollments resulted in “W” grades (withdrawals). The impact of receiving a large number of “W” grades is quite negative. By contrast those TVP students with a moderate to low level of “W” grades (fewer than 20% of grades are “W” grades) were much more likely to successfully transfer (59%) than those students with more “W” grades. There is an inherent tension between “W” grades and GPA. Students who withdraw from a course may receive a “W” but those who stay even though they may have missed a large number of classes or perhaps are simply not “getting it” will receive an “F” or “D” grade (in all likelihood). In other words, it is a dilemma for the student and for those who would advise that student. In most cases where a “W” grade is received the only other possible option would have been an “F.” Neither are desirable outcomes, but at least the “W” grade allows for several opportunities to re-take the class with no negative impact on the student’s cumulative GPA. Used judiciously, the “W” option may help students, but in excess, it is not going to facilitate student transfer. It seems that in this case interventions aimed at keeping a student focused on schoolwork and attending class regularly may be the best hope of reducing the incidence of both “W” grades and non-passing grades. Enrollment Patterns Several enrollment patterns are positively associated with transfer: 1) attending a CCC while in high school (i.e., concurrent enrollment); 2) attending more than one community college; and 3) enrolling in summer sessions within the first three years of college. Of these three, summer enrollment has the largest relative risk (47%) and it also has the virtue of being a relatively straightforward matter to encourage both through policy and through direct action. Concurrent enrollment is modestly transfer-enhancing factor, suggesting that encouraging high school students to take college courses while they are still attending high school will predispose those students to be more successful at the community college, or perhaps to enter directly into a four-year school. In any case, students who were concurrently enrolled at the high school and at the community college were 11% more likely to transfer than those students who did not experience concurrent enrollment. Enrollment at multiple colleges has a positive relative risk; students who attend multiple colleges are 15% more likely to transfer than students who attend only a single college. Forty percent of TVP students have enrollment records at more than one college. The reasons the rate of multiple college enrollment is so high are many. For example, students may change residences with the result that another college is more proximate; one college may offer courses that are not full or not offered at the student’s primary college; a different college may have a better reputation or may offer different services, and so on. Enrollment at multiple colleges is common in the TVP cohort; but why does it have a positive effect on transfer success? Reasonable arguments can be made for anticipating either a positive or negative effect of attending multiple community colleges. On one hand, when a student changes colleges it may be disruptive, requiring adaptation and extra effort, such as more time commuting, new teachers, and so on. These disruptive effects might sap resources and rob attention that could better be devoted to study. On the other hand, attending multiple colleges may allow students to expedite their progress. They can pick up classes that are impacted or unavailable at their primary institution and gain practice in navigating a new institution. They may need to transfer transcripts, credits, or test scores – activities that may be necessary to waive pre-requisites or to take an advanced class. In addition to the increase in transfer velocity that can be achieved by picking up needed classes, navigating the bureaucracy of another institution may prepare students for the transfer experience with a four-year school. Multiple community college attendance was predictive of transfer in the endogenous college-level model, as well, suggesting that this finding is robust. To further test the assumption that student swirl is appropriate to consider as an endogenous factor, we analyzed the distribution of student swirl to determine if swirl were purely a function of geographic location. We found that there was no association between college size (a proxy for rural/urban) and the proportion of students with multiple community college attendance. Breaking the colleges into quartiles based on the average swirl level of their student bodies showed that there were rural colleges with low swirl levels, but that there were also rural colleges with intermediate and high swirl levels. Finally, we observed a moderately strong correlation between average swirl level and the BA Plus Index (r = .47) indicating that lateral transfer or swirl may be a “proximal” variable that partially explains how the BA Plus Index creates its effect. In other words, lateral transfer may be a strategy that students from areas with high average levels of education use to advance their educational aspirations. time frame. On the other hand, 15% of TVP students took neither a math nor a reading class in their first year at the community college. Students who showed this pattern of postponing enrollment in both math and English were one third less likely to transfer than those who took math and/or English in their first year. In general, students who start at transfer level enjoy better odds of successful transfer, than students who start at the degree-applicable, non-transferrable level or at the basic skills level. Students who start at the degreeapplicable, non-transferrable level of English are 10.4% less likely to transfer than those who start at the transfer level; those who start at the basic skills level are 12.9% less likely to transfer than those who start at the transfer level. The relative risk statistics for initial math level are more pronounced than those for initial English level. Students who start at the degree-applicable, non-transferrable level in math are 32.4% less likely to transfer than those students whose initial math class is at the transfer level. Those starting at the basic skills level in math are experience a 48% reduction in their odds of transfer, relative to those who start at the transfer level. Not taking any math at the CCC resulted in a 51.4% reduction in the odds of transferring, indicating that math is an especially difficult bottleneck for many students. Working closely with high schools to improve the initial placement levels of incoming students should be an important aspect of any plan to improve the number of Bachelor’s degrees awarded in California. Students’ enrollments in Career Technical Education (CTE) classes were also found to be associated with transfer. If 10% or more of a student’s units are from Career Technical Education (CTE) classes, the likelihood of transfer is decreased; students with transcripts containing fewer than 10% of CTE enrollments were 14% more likely to transfer than those students whose transcripts contained more CTE enrollments. The three categories of “extra assistance” classes (Guidance, Reading, and Study Skills) all had low incidences of use among the TVP. They also had small negative associations with transfer (except for study skills, which had no significant effect), most likely because those students who did seek out these classes were seeking to overcome academic weaknesses. Another possibility is that the guidance and/or reading classes were being used disproportionately by job-seeking students who have a lower incidence of transfer. Course-Taking Patterns Degrees and Certificates Previous research (e.g., Adelman, 2005) has shown that taking math and English courses immediately upon college entry is correlated with degree completion and transfer. In the TVP we found that 88% of students took at least one math class within six years of enrolling and that 91% of TVP students took an English class in the same There is a very large positive effect of obtaining an Associate’s degree (either an AA or AS) within either the first six years of enrollment. Obtaining an AA/AS was found to be a time variant factor in relation to transfer. That is, the effect is not constant over time, it is more pronounced for those students who attain a degree within the Vol. 18, No. 2, Spring 2011 / 29 first three years, or “early”; these students were 103% more likely to transfer than students who did not receive an AA within the first three years. Students who attain a degree in the middle years (four through six) still enjoy a robust positive effect (49% more likely to transfer). It may be that California’s road to more Bachelor’s degrees is actually paved with Associate’s degrees, particularly for African American and Latino students who experienced a much greater boost from attaining an Associate’s degree (193% and 263%, respectively) than did whites and Asians. By contrast, the 6% of TVP students who attain certificates are 40% less likely to transfer than those who do not receive certificates. The effect does not vary over time; it is negative regardless of the year in which the certificate is attained. It is likely that many of the students who obtain certificates are “false positives” in terms of their entry into the TVP cohort. That is, many of these students have no intention to transfer but instead are seeking a terminal certificate and employment in a related field. Priorities to Consider It is very important to realize that a large data mining project such as this that is based on correlational methods does not directly support causal inferences. It is difficult to rule out competing hypotheses and draw truly causal conclusions. Still, efforts were made to include theoretically sound variables and relationships which might suggest action. Additional data sources, qualitative assessment and future research will be able to tease out which relationships are truly causal. Action based on findings from the current study could be evaluated to judge the causality of the relationships suggested here. While a great deal of information was analyzed in the TVP, certain findings rise to the top in terms of the strength of their associations and their accessibility to intervention. We consider several of these now. Class Withdrawal Withdrawals are a significant issue that should be tackled head-on by the system as a whole. The impact of many “W” grades is clearly detrimental to transfer. While there is evidence that excessive “W” grades are negatively associated with transfer, this is an area where causal interpretations must be carefully considered. On campuses, the “W” grade is often seen as a student-friendly way of letting students who would otherwise fail have a chance at being successful in their next attempt. While excessive use of the “W” option is not associated with student success, the alternative grade is usually an “F” and a low GPA does not promote transfer either. Both “F” and “W” grades are likely to stem from the same persistent causes (e.g., inadequate skills, disruptive personal issues, lack of commitment, etc.). Limiting or prohibiting “W” grades 30 / Journal of Applied Research in the Community College would be a bit like outlawing aspirin in an effort to stop people from getting headaches. It would be an intervention aimed at the symptom, not the cause. Proper intervention requires reflection upon root causes. Effective actions to combat “W” grades are likely to involve new resources and the reconfiguration of existing resources. For instance, students who are at risk for failure could be identified at the time of enrollment based on diagnostic information that is already available in their transcript or from their placement testing. They could then be routed to special class sections such as learning communities which provide greater support and guidance to students who need more personal support in order to succeed. Such a proactive approach to heading off “W” grades before they occur is analogous to the medical practice of screening for risk factors. When risk factors are identified (e.g., a student presents with a history of “W” grades) then resources could be targeted to advise, track and if necessary offer tutoring, special assistance, legal advice, coaching, or remediation as the situation demands. Summer Enrollment The student-level model suggests that students should be encouraged to enroll in summer terms, particularly in their first three years. The observed effect may be partly due to the self-selection of more motivated students choosing to enroll in summer school, but there is good theoretical basis for expecting summer enrollment to increase student engagement and retention, ultimately facilitating transfer. Math Support In terms of supporting specific course work that will enhance the odds of student transfer, the greatest leverage may be gained by adding additional supports and processes that will ensure that students are capable of attaining and completing transfer-level math. To this end, greater alignment and collaboration between high schools and community colleges may increase the level of initial student placement and thereby enhance entering students’ chances of transfer success. Further pedagogical innovations should be actively explored and supported as the system seeks ways to improve math success. Encourage Associate Degree Attainment Finally, students should be encouraged to attain an associate’s degree as quickly as possible upon entry to the CCC even if they express a desire to transfer. Far from being a distraction, the attainment of an associate’s degree is one of the most positive, transfer-facilitating factors in the study. Attaining an AA or AS within the first three years of enrollment increases the odds of transferring by 103%. Part of the reason for the strong impact is that the qualifying for an AA or AS within three years implies a fairly consistent level of mostly full-time enrollment. But beyond the accumulation of units, qualifying for an associate’s degree requires a certain transfer-conducive structure to the units acquired. Thus, students’ with associate’s degrees may not stray far from transfer requirements and have maintained a good transfer velocity. Associate’s degrees do not fully map to transfer, of course, and the requirements vary by CCC and by transfer destination, but in many key respects the AA/AS pathway provides guiderails for students to maintain their transfer velocity. Future Research Future research should extend the current research findings and method to an analysis of those factors that lead first time freshmen to pass the threshold of entry into a TVP cohort (or not). Students in the TVP had a relatively high transfer rate, accounting for 67% of transfers from a given cohort year of all entering first-time freshmen, despite representing only 25% of the larger first-time freshman cohort. It is possible that the return on investment by encouraging more students to get to the initial momentum point of entering a TVP cohort (i.e., attempting a transfer level English or math course and accumulating 12 for-credit units) will be larger than focusing efforts on raising the transfer success of those students in the TVP, an already largely successful group. Applied research into the transfer function would benefit from the development of a new transfer cohort possessing characteristics that would enhance its usefulness to researchers, practitioners, and policy makers. Work should begin to create a transfer cohort that can be identified at three years, rather than six. Our research indicates that transfer patterns for a cohort of first-time freshmen are well-set by the third year. A shorter time frame to set a cohort would allow for more rapid analysis and feedback in response to initiatives and programs that support transfer. Transfer rates become reliable at three years and can be used to predict six year and twelve year rates with a high degree of accuracy. Any future transfer cohort definition should have at least the same level of sensitivity and specificity that is seen in the current TVP cohort definition. A comprehensive study of the impacts of financial aid on transfer velocity would be a complex but very worthwhile undertaking. The apparent under-utilization of financial aid by certain ethnic groups in the TVP cohort, most notably Filipinos and Whites, provokes more questions, perhaps, than it answers. In an effort to cross-check the pattern observed in the 1999-2000 Transfer Cohort, we examined the entire system enrollment for the 2000-2001 academic year and found a pattern of financial aid usage across the entire student population quite similar to that observed in the TVP. The similar patterns of financial aid usage suggests that more outreach should be done in order to encourage greater use of grants and Board of Governor (BOG) fee waivers, which do not have to be repaid. And yet, there is still much that remains to be explored in terms of the effectiveness of financial aid. Recently, a new data element that assesses first-generation college status has been added to the centralized CCC application program. In the future this data element will be useful in providing a statistical control for all students. In the past, there was little statistical information available about students who did not choose to apply for financial aid. Can we recommend greater use of loans and work study? Does the relatively modest size of the positive relative risk warrant widespread adoption of these forms of student aid? There is some evidence that loans and work study are particularly effective forms of financial aid for promoting student transfer and yet their utilization rates are quite low. This finding is important because little work has been done at the Community College on the relative merits of student loans versus other types of aid. Some researchers have reported finding a negative to neutral impact student loans on student success at four-year colleges (Doud & Coury, 2006; Kim, 2007). Prior research on the effects of the form of financial aid known as work study has found that work study is associated with higher student retention at college. This effect is typically explained as a result of higher levels of student engagement (e.g., Astin, 1999). The TVP found that student participation in work-study is a positive, transfer-facilitating factor that increases the odds of transfer by 13%. Kane (2003) found that CalGrants had a strong effect on initial college-going behavior. The current study found that receipt of non-Pell grants (i.e., “Other Grant”) – the vast majority of which are CalGrants – increased the odds of transfer by 8%. This is another important TVP finding concerning the effects of financial aid. Because of the wide ethnic differences in financial aid utilization observed in this study and others (e.g., Kane, 2003) a more extensive analysis of the impacts of financial aid across various ethnic and socioeconomic groups is called for. Studying the combinatorial and interaction effects of different patterns of aid utilization over time would deepen our understanding of the true impact of financial aid and allow for better optimization of resources and results. Another area worthy of additional research is multiple community college attendance. A more in-depth investigation of lateral transfer and its benefits for students would shed more light on a common phenomenon that is beginning to emerge as a way of empowering students to advance their academic career and course-taking agenda. How can multiple-college enrollment be managed to enhance student success in larger numbers? In closing, the goal of interventions targeting transfer Vol. 18, No. 2, Spring 2011 / 31 success should be to increase the transfer velocity students flowing from community colleges into institutions where they can attain a Bachelor’s degree. The effect of increased efficiencies resulting in reducing the average time to transfer for the TVP cohort by one year would result in overall 4% increase in the standard six-year transfer rate, representing 6,000 more transfers annually. The cascade effect of having all six-year transfers complete in five years and all five-year transfers complete in four years, etc. would create a large swell in the transfer-ready population. No one intervention is likely to result in such a potent effect, but a series of well-targeted, well-resourced interventions, calibrated by accurate assessment of the transfer function would greatly increase the odds. References Adelman, C. (2006). The Toolbox Revisited: Paths to Degree Completion From High School Through College. Washington, D.C.: U.S. Department of Education. Attewell, Paul, Lavin, David, Domina, Thurston, and Levey, Tania (2007). Passing the Torch: Does Higher Education for the Disadvantaged Pay Off Across the Generations? Russell Sage Foundation Bailey, T., Jenkins, D., & Leinbach, D.T. (2006). Is Student Success Labeled Institutional Failure? Student Goals and Graduation Rates in the Accountability Debate at Community Colleges (Working Paper No. 1). New York: Community College Research Center, Teachers College, Columbia University California Community College Chancellor’s Office (2007). Transfer Rate Study of California Community Colleges (2005-06 Report), available at www.cccco.edu/ Portals/4/TRIS/research/reports/transfer_report.pdf. California Department of Finance (2006). California Public Postsecondary Enrollment Projections, 2006 Series, Sacramento, California. California Postsecondary Education Commission (2007a). California Higher Education Accountability: Goal—Student Success, Measure: California Community College Students’ Degrees and Certificates Awarded and Successful Transfers, available at www.cpec.ca.gov/ completereports/2007reports/07-06.pdf. California Postsecondary Education Commission (2007b), College-Going Rates: A Performance Measure in California’s Higher Education Accountability Framework, Commission Report 07-04, available at: www.cpec.ca.gov/completereports/2007reports/07-04.pdf California Postsecondary Education Commission (2008). Who Can Afford It? How RisingCosts Are Making College Unaffordable for Working Families, Sacramento, California. Campaign for College Opportunity, Practices with Promise 2008: A Collection of Working Solutions for College Opportunity and Student Success, available at www.collegecampaign.org/assets/docs/ pwp/pwp-full-report-2008-11-19.pdf. Dowd, Alicia and Coury, Tarek (2006). “The Effect of Loans on the Persistence and Attainment of Community College Students”. Journal Research in Higher Education (Vol. 47, No.1, pp. 33-62). Driscoll, Anne K., “Beyond Access: How the First Semester Matters for Community College Students’ Aspirations and Persistence,” Policy Analysis for California Education, Policy Brief 07-2, Berkeley, CA, August 2007. Hill, Elizabeth G., Back to Basics: Improving College Readiness of Community College Students, Legislative Analyst’s Office, Sacramento, CA, 2008. 32 / Journal of Applied Research in the Community College Hom, Willard (2009). “The denominator as the target,” Community College Review, 37 (2) Horn, L., & Lew, S. (2007a). “California Community College transfer rates: Who is counted makes a difference,” MPR Associates, Inc., Berkeley, CA. Horn, L. & Lew, S. (2007b). “Unexpected pathways: Transfer patterns of California community college students,” MPR Associates, Inc., Berkeley, CA Hans, J. & Reed, D (2007). “Can California import enough college graduates to meet workforce needs?” California Counts, Vol. 8, No. 4, Public Policy Institute of California, San Francisco, CA, May 2007. IPEDS (2010). Accessed on 8/7/10 https://surveys.nces.ed.gov/ipeds/ VisInstructions.aspx?survey=4&id=491&show=all Kane, Thomas J. (1994). “College entry by Blacks since 1970: The role of college costs, family background, and the returns to education,” The Journal of Political Economy, 102 (5), pp. 878–911. Kane, Thomas J. (2003). “A Quasi-Experimental Estimate of the Impact of Financial Aid on College-Going,” National Bureau of Economic Research Working Paper 9703, available at www.nber.org/papers/ w9703.pdf. Kelly, Patrick J., As America Becomes More Diverse: The Impact of State Higher Education Inequality, National Center for Higher Education Management Systems, November 2005, available at www.nchems. org/pubs/docs/Inequality%20Paper%20Jan2006.pdf. Kim, D. (2007). “The Effect of Loans on Students’ Degree Attainment: Differences by Student and Institutional Characteristics”. Harvard Educational Review, 77 (1). Mery, P. & Schiorring, E. (2008). “A Qualitative Study of Two-to-FourYear Transfer Practices in California Community Colleges,” RP Group, available at: www.rpgroup.org/documents/TLC_Cross_ Case_Analysis.pdf. National Center for Public Policy and Higher Education (2008). Measuring Up 2008: The National Report Card on Higher Education, San Jose, CA. Offenstein, J., Moore, C., & Shulock, N. (2010). Advancing by Degrees: A Framework for Increasing College Completion, Institute for Higher Education Leadership & Policy and The Education Trust, Sacramento, CA. Provasnik, S., and Planty, M. (2008). Community Colleges: Special Supplement to The Condition of Education 2008 (NCES 2008-033). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC. Reed, Deborah, California’s Future Workforce: Will There Be Enough College Graduates? Public Policy Institute of California, San Francisco, CA, 2008. Rostgaard, Klaus (2008). “Methods for stratification of person-time and events – a prerequisite for Poisson regression and SIR estimation,” Epidemiological Perspectives & Innovations, 5(7) available at: http:// www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2615420 Sengupta, R., and Jepsen, C. (2006). California’s Community College Students, California Counts, Vol. 8, No. 2, Public Policy Institute of California, San Francisco, CA. Shulock, Nancy, and Colleen Moore, Rules of the Game: How State Policy Creates Barriers to Degree Completion and Impedes Student Success in the California Community Colleges, Institute for Higher Education Leadership and Policy, Sacramento, CA, 2007. Craig Hayward, Ph.D., is Director of Planning, Research, and Knowledge Systems at Cabrillo College