The Jackson Hole Higher Education Group, Inc.
CyberCampus Project
Technical Document 3.1
The Simulation Engine: Initialization
This paper describes work in progress on the Higher Education simulation project funded by the Alfred P. Sloan Foundation. Contents may not be used or cited without permission. Limited distribution is provided to obtain comments and criticisms, and to assist potential development partners. Copyright © 1998 by The Jackson Hole Higher Education Group, Inc..
1. Introduction
2. Database Variables
3. Player Input
4. Measures of Prestige and Similarity
4.1 Institutional Prestige
4.2 Similarity to the PGI Specification
4.3 Screening and Descriptor Variables
5. Initial Conditions Variables
5.1 Transformation S-Curves
5.2 Definitions: Derived Variables
6. Peer Institution Benchmarking Variables
7. Undergraduate Applications and Yield Rates
7.1 The Student Preference Function
7.2 PGI Applications and Yields
Endnotes:
Higher Education is a computer-based simulation game under development that targets both the institutional professional and the interested layperson to participate in leadership challenges in a college or university setting. Players set, monitor, and modify a variety of institutional parameters and policies, allocate resources as they see fit, and watch as results continually unfold. The game provides an opportunity to experiment and succeed or fail in a safe and entertaining fantasy environment. While Higher Education is necessarily a caricature of real academic life, it is grounded in authentic data and will provide serious lessons in higher education. The game will be driven by a sophisticated simulation engine that models six broad areas:
1. enrollment management
6. initialization procedures
Models 1-5 are described in Technical Documents 2.1 through 2.5, prepared during the project's preliminary design and prototyping phase. An overview is provided in Technical Document 2.0. The engine's development was supported by the Sloan Foundation and the Spencer Foundation.
This paper describes the data and associated models used in the game's initialization: specifically, the determination of initial conditions and the peer institution benchmarking variables for the player-generated institution. Initial conditions that cannot be derived from the available datasets will be developed at a subsequent design stage .
Data for the CyberCampus project have been provided by the University of Pennsylvania's Institute for Research on Higher Education (IRHE) under a grant from the Sloan Foundation. The initialization dataset consists mainly of IPEDS data, supplemented by data from the College Board and other sources as needed for particular variables. The variables will be defined as they are introduced below. We have transformed the data as needed to fit the CyberCampus specifications. The transformed dataset will be stored on the CyberCampus CD-ROM and used as described herein.
The initialization data are used to calculate an index of institutional prestige and a measure of the similarity between the player-generated institutional specifications and each institution in the dataset. Next the game identifies the n institutions (n=10 in the prototype) that are most similar to the player's specification, then calculates the initial conditions variables for each of them. The initial conditions variables for the player-generated institution equal the averages of the variables for the n most similar institutions.
A final procedure identifies a list of benchmark institutions whose similarities are greater than a threshold value, sorts them by the prestige index, and then calculates a set of benchmarking variables for these schools. The benchmarking variables are also calculated for the player-generated institution, which is inserted into its proper place the benchmarking tableau.
We have not yet decided whether to identify the benchmark institutions by name, although this appears feasible given that the data are in the public domain. To do so might add player interest, but it might also raise questions about whether particular schools ought to be in the benchmarking set.
The design of a game like CyberCampus, which combines complexity and aspirations for realism, presents an important challenge. On the one hand, the game's startup process must be simple enough to allow the player to move quickly into actual play. Yet the player must be able determine the kind of institution he or she wants to manage. Selecting the kind of institution requires a significant amount of data, either from the player or from the game's database. The data must be internally consistent and it must reflect the desired institutional type. The need to maintain player interest and the difficulty of inventing mutually consistent data elements led us to minimize the amount of input required. Instead of asking the player to provide voluminous data, the game will rely on an internal file of actual institutional data. The algorithms described herein will extract the data most representative of the kind of institution the player wants to manage.
Figure 1 shows the kind of information the player will provide, plus certain parameters and calculations used by the algorithms. The player's inputs are arrayed down the left side of the figure, the parameters are on the right, and certain intermediate calculations are in between. The individual fields will be explained in context as we describe the algorithms. In broad terms, though, the player specifies whether his or her school will be national, regional, or local in scope, its general location (important for regional and local schools), campus environment, size, prestige, certain descriptors of program mix, and an indication of financial strength.
The player's inputs drive the similarity calculations according to a procedure to be described presently. First, though, we will describe the computation of institution prestige, upon which the similarities depend.
The prestige index simulates the rankings found in U. S. News and World Reports and similar sources. While we do not believe such indices represent true measures of institutional quality, they do reflect the market's perception of traditional academic prowess. Moreover, the CyberCampus algorithms require an overall market index which we may as well call "prestige."
CyberCampus defines prestige as a function of four undergraduate admissions variables and four general institutional descriptors:
The variables are normalized by setting the sign so larger is better and then dividing by the standard deviation. The prestige index is calculated by taking the weighted sum of the normalized variables. Final weights will be set after alpha testing.
The distance function quantifies the degree of similarity between the player generated institution (PGI) and each of the 1200 institutions in the game's database. "Distance" equals the weighted sum of squares of the deviations of the individual variables from their targets. (This is the square of the traditional Euclidean distance measure.) Smaller distances imply greater similarity.
The following variables are used in the distance function. Their values are provided by the player as shown in Figure 1 unless otherwise noted.
1. Institutional prestige: defined above
2. Number of students: total student headcount
3. Percent of undergraduates who are part-time
4. Percent of undergraduates getting the bachelors degree in five years: 0-100 scale defined in terms of percents (a measure of how traditional is the student body)
5. Percent of full-time undergraduates in student housing: 0-100 scale, applies to Student Level 1 only
6. Athletic program intensity: a 1-5 scale, defined in terms of NCAA Division status, as follows:
=5: I-AA football
=4: Other Division I football and/or Division I basketball
=3: Division II football or basketball
=2: Division III football or basketball
=1: No football or basketball division listed
The player input "Athletics program intensity" is defined in terms of the same 1-5 scale.
7. Percent graduate students: percent of student headcount pursuing masters or doctors degrees (SL 3 & SL 4).
8. Percent non-degree students: percent of student headcount not pursuing a degree (SL 5).
9. Doctoral program intensity: a 0-10 scale (defined later) based on the number of doctoral degrees per tenure-line faculty member.
10. Sponsored research intensity: a 0-10 scale (defined later) based on sponsored research dollars per tenure-line faculty member.
11. Institutional wealth rating: a 1-3 scale tentatively defined as follows.
1. Rich: (EpS>50.53 or EpF>271 or FBpOE>0.147) and TpOE<0.32
2. In between: neither 1 nor 3
where:
EpS = endowment per student; EpF = endowment per tenure-line faculty member; FBpOE = fund balances as a percent of operating expenditures; TpOE = tuition revenue as a percent of operating expenditures;
and the breakpoints represent the 25th and 75th percentiles of the individual variables for the 1,200 sample institutions.
Three additional variables are used in setting up the game's initial conditions, benchmarks, and campus environment.
1. Institutional control: "public" or "private." CyberCampus limits institutional selection for calculation of initial conditions to schools with the same institutional control as the PGI. This is required because some of the financial and pricing variables depend critically on the type of control. However, the peer institution set (which deals mainly with institutional performance and student selectivity) may include both institutional types.
2. Geography: players specify the region of the country in which the PGI will be located and whether it will appeal to undergraduate students on a national, regional, or local basis.
• National: geography doesn't affect the distance measure; region matters only in that it may affect the art associated with campus environment and seasonal changes.
• Regional: only the schools in the states tentatively shown, for each region, in Figure 2 can be candidates for the initial condition and peer institution set: in effect, the distance for institutions in other states is set to infinity.
• Local: schools must meet the regional criteria. For those that do, an additional distance factor will be applied based the following algorithm: (i) draw a random number to determine the PGI ‘s state from the set specified in Figure 2 for the region; (2) determine the factor from a table that shows the relative student market drawing power for each pair of states on a 0-3 scale. High drawing power will decrease distance, lower drawing power will increase it, and zero drawing power will make the institution ineligible.
3. Campus environment: "urban", "suburban", and "rural": this variable determines the art that describing the campus locale and may affect student demand, crime, and other factors, but it does not affect the initial conditions or peer benchmarks.
The initial conditions model provides starting data elements that are mutually consistent an consonant with the player's specifications. As many data elements as possible are based on data from actual institutions contained in the CyberCampus database. Additional data elements will have to be obtained by judgment. This paper deals only with the variables derived from the actual data.
Figure 2 lists the initial conditions variables. Each variable is obtained by averaging the data for the 10 institutions of the same control type that are most similar to the desired PGI. The results of this calculation should be displaced by a random variable to avoid perfect predictability with a given set of player specifications. The variable definitions follow those used in IPEDS and other standard sources except as noted below. The non-standard variables are derived from the underlying data elements as described herein. However, the game database will contain the derived variable only, not the underling data elements.
We emphasize that the purpose of the initial conditions calculations is to obtain a plausible and internally consistent set of starting values that generally reflects the player's wishes, not to reflect the situation at any real-world institution. The player will never know which institutions were used to generate the initial conditions. Hence we are able to manipulate the data in ways that would not be appropriate for substantive statistical research.
Preliminary analysis revealed the need to transform the data using s-shaped functions in certain cases where the raw variables cannot be used. (The cases will be described below.) An example of such a function follows:
S-Shaped Transformation Function
The s-function transforms the input variable (on the horizontal axis) into a new variable (the vertical axis) with a new dimensionality and a pre-specified minimum and maximum. These properties are particularly helpful when one wants to control a variable's range without imposing discontinuities at the upper and lower bounds.
The s-function requires five parameters, which we define as the upper and lower bounds (upB, loB), the midpoint (mX) and a pair of x- and y-values that determine the curve's slope (sX and sY). The midpoint is the input value which produces an output half-way between the upper and lower bounds: mX=0.5 in the figure, mY is always preset at (upB–loB)/2. The slope values are taken as deviations from the midpoint values of x and y: in the example, {slX,slY }={.05,.05} would mean that sX and sY both equal 0.5+0.05 = 0.55. The s-curve equation is:
(1)
The inputs for x and y determine a and b.
The derived variables are designated by a check (√) in the "Derived?" column of Figure 2. The definitions for the other variables are the same as in the source databases and are not repeated here. The assumptions used for preprocessing are preliminary and can be changed easily. While some of the assumptions can be researched, time limitations have prevented us from doing so. Advice and research will be appreciated.
1. Replacement value of buildings: we have the book value of buildings, but CyberCampus requires replacement cost for its construction and renovation algorithms. The conversion procedure relies on rough assumptions about the time-frame of construction during the last thirty years (5 percent of the prior year's replacement value added each year), inflation (varies year by year)m real construction cost-rise (3% per year), and depreciation (2.5% per year). Given a starting value of 1, these assumptions produce a book value of 16.8 and a replacement value of 35.8 after thirty years, for a conversion multiplier of 2.1.
2. General plant debt and residence hall debt: we have data on total debt, whereas CyberCampus requires debt to be broken down into the general and residence hall components. We assume that total debt includes residence hall debt, make a preliminary of estimate residence hall as a percentage of total debt, and then transform the preliminary estimate using the following s-shaped curve:
input:
output: percent residence hall debt
The preliminary estimate of residence hall debt depends on some rough assumptions about construction cost per student space ($45,000 in 1997 dollars), the percentage debt financed (90%), the debt amortization rate (3.3% per year), and the time frame of construction (spread equally over the last thirty years). Transformation through the s-shaped curve is required to ensure that the fraction of total debt due to residences remains between zero and one.
3. Sponsored research revenue: equals the sum of the IPEDS figures for restricted and unrestricted research revenue. Government-owned contractor operated (GOCO) facilities like the Stanford Linear Accelerator Center are included in the data but logically excluded from the game. However, this affects only a small number of institutions.
4. Gifts for current operations: our database includes total restricted and unrestricted gifts, but it does not provide the fractions restricted to endowment or plant, or sums transferred to quasi-endowment. Because CyberCampus's "gifts for current operations" variable counts as part the operating revenue-expense balance, we must make an assumption about what percentage of total gifts is used for current purposes. Again we use an s-shaped curve:
input: total gifts as a fraction of total operating expenditures
5. Endowment spending rate: while our data include endowment spending and market value, their ratio does not necessarily reflect the institution's spending rate policy. Recent above-average investment returns and low dividend yields tend to depress realized payout rates. Institutions occasionally withdraw lump sums for special purposes, and these are included in "spending from endowment." Because the CyberCampus initial financial statements should reflect long-term spending policy, we transformed the computed spending using the following s-curve:
input: reported endowment spending/market value
6. Athletics revenue and expense: we have data on schools' intercollegiate athletics programs, but not on costs or revenues. We assume that schools with high athletics ratings generate gate and TV revenues and incur costs that are independent of student numbers, whereas all schools incur costs for recreational programs that are proportional to student numbers. (Recreational programs produce no revenue.) We make the rough assumption that 60% of athletics cost represents non-salary expense.
Figure 4 presents preliminary assumptions about intercollegiate base costs and revenues. ("Base" means the figures represent a 50-50 won-lost record; the revenue figures should be scaled upward or downward by perhaps one-third in the case of 75% and 25% wins, respectively.) The model determines the NCAA division for football and basketball (stored in the game's database) and adds up the appropriate entries from the table.
Figure 5 presents preliminary assumptions about the cost per student for the various student levels and institutional segments. The cost coefficient are applied to the enrollment data to complete the expense side of the athletics equation.
7. Interest on Operating Reserve: equals the operating reserve balance time a notional short-term interest rate (input as 8%, probably too high).
8. Departmental Faculty Salaries our database includes total departmental salaries, but it does not divide them between faculty and staff as required by CyberCampus. Therefore, we project faculty salaries from database values for faculty numbers and average salary by regular faculty rank. Average salaries for adjuncts are taken as a weighted average of the regular faculty ranks, using the weights shown in Figure 6.
The projection for total faculty salaries is divided by the database figure for total salaries to get a preliminary estimate of the faculty salary fraction. Because the numerator and denominator come from different sources, the ratio may be unreasonable in some cases. Therefore, we pass the preliminary estimate through the following s-shaped curve to guarantee the result will be reasonable:
input: projected faculty salaries as a fraction of IPEDS total salaries
9. Sponsored Research Expenditures and Overhead Rate: we have data on sponsored research salaries and non-salary expense (which sum to total research revenue). The first task is to separate the non-salary figure into direct and indirect components: i.e., between expenditures by principal investigators for supplies, equipment, travel, etc., and overhead payments to the university. Then we must separate the salaries component into academic-year faculty salary offsets and all other salaries. (We simplify by including faculty summer salaries under "other salaries.")
We assume that the institution's effective overhead rate is an s-function of its total research volume. ("Effective" means the average rate actually charged as opposed to the negotiated rate.) The justification is that more research-intensive institutions put more effort into rate calculation and negotiation and that they enjoy more market power (and thus required fewer discounts from the negotiated rate) than less research-intensive schools. The parameters of the s-curve are as follows:
input: total sponsored research volume ($millions)
output: effective indirect cost rate
The initial conditions figure for other research expense is taken net of overhead. Overhead is included in sponsored research revenue (actually, unrestricted sponsored research revenue). However, to include it on the expense side would double-count support service expenditures funded by the overhead collections.
We assume that academic-year faculty salary offsets are a function of research volume per faculty member. The justification again lies in the distribution of market power. The parameters of the s-shaped curve follow:
input: total sponsored research per faculty member ($thousands)
output: academic-year faculty salary offsets as a fraction of total research salaries
10. Institutional Advancement Expenditures: our data do not break out institutional advancement (fund-raising) from institutional support. We estimate it as an s-function of total gift receipts. The parameters of the function are:
input: total gifts ($millions)
output: institutional advancement expense ($millions)
parameters: {loB, upB}={.5,10}, mX=100, and {sX,sY}={150,8}.
The resulting sum is prorated between salary and non-salary expenditures, with a ceiling of 25% of each category imposed on the output of the s-curve, and the result removed from institutional support.
11. Academic Information Technology Expenditures: our data do not break out information technology from academic support expenditures. Having identified no plausible driver variables, we simply took 10% of academic support salaries and 30% of non-salary expenditures as pertaining to information technology.
12. Service on General Plant Debt: equals a notional interest rate for tax-exempt long-term debt (input as 6.5%).
13. Surplus-Deficit: while an institution's "raw" surplus or deficit can be computed from the data on revenue and expense, the result is not reliable enough to drive the CyberCampus initial conditions. The revenue and expenditure figures are not fully comparable, and short-term factors also can introduce significant noise. Therefore, we smooth the surplus-deficit figure by passing it through the following s-curve:
input: raw surplus-deficit as a fraction of total operating expenditures
output: smoothed surplus-deficit as a fraction of total operating expenditures
parameters: {loB, upB}={–.03,.03}, mX= –.01, and {sX,sY}={–.04,–.02}.
The figure for raw surplus-deficit includes interest on the current operating reserve and the cost of servicing plant debt. An institution's surplus-deficit is obtained by multiplying the output by its total operating expenditures.
14. Transfer to Plant: we have no data on the sums, if any, that institutions transfer into their capital reserve. (We do have the capital reserve balance, also known as unexpended plant funds.) Yet CyberCampus needs to show a transfer to plant in its initial financial statements. Because it seems logical that the transfer bears some relationship to replacement cost depreciation and to the institution's current surplus-deficit, we obtain the needed figure by means of the following s-curve:
output: transfer to plant as a fraction of plant replacement value
An institution's transfer to plant is obtained by multiplying the output by the product of its plant replacement value and a notional depreciation rate (0.025). For example, a school with a very large surplus would transfer half of its computed replacement cost depreciation (upB=.5).
15. Other operating income: balancing the income-expense equation in light of the surplus-deficit adjustment and transfer to plant requires an offsetting adjustment to some other variable. We chose other operating income. For CyberCampus,
other operating income = raw other operating income + surplus-deficit + transfer to plant – raw surplus-deficit
Note: we may wish to revisit this algorithm because the resulting value for other operating income is negative in some cases, and negative values would appear implausible.
16. Doctoral Degrees per Faculty Rating: equals doctoral degrees per tenure-line faculty member, converted to a 0-10 scale based on the range of the computed ratio.
17. Sponsored Research per Faculty and Sponsored Research Rating: research per faculty equals total sponsored research revenue per tenure-line faculty member; the rating equals research per faculty converted to a 0-10 scale.
18. Masters and Doctoral Enrollments: our data combine the enrollments for masters and doctoral students, but CyberCampus needs the disaggregated figures. We first estimate doctoral enrollments as the minimum of (a) 40% of graduate and professional enrollment and (b) the product of the number of doctoral degrees awarded and an assumed value (3.5 years) for the average doctoral student lifetime in residence (i.e., the time from matriculation to departure with or without a degree). Masters enrollment is obtained by subtracting doctoral enrollment from graduate and professional enrollment.
19. Undergraduate Target Student Intake: our data combine the intake data for traditional (FT) and non-traditional (PT) undergraduates (SL-1 and SL-2) , but CyberCampus needs the disaggregated figures. (We refer to this as a target because actual intake may differ due to fluctuations in the yield rate.) We also have the total enrollments of full-time (SL-1) and part-time (SL-2) students. Dividing by the appropriate assumed lifetime (input as 4.2 years for traditional, 7.0 years for non-traditional students) produces a first-approximation to the intakes. We obtain our final estimate by prorating the actual enrollments according to the first-order estimates:
(2)
20. Graduate and Non-matriculated Target Student Intake: we have no intake data for SL-3 through SL-5. Estimates are obtained by dividing the enrollment figures by the relevant student lifetimes.
21. Undergraduate "Calibration" Graduation Rates: the CyberCampus academic operations model determines the number of graduating students, and hence the graduation rates, on the bases of degree requirements met. However, the initial conditions data help calibrate the academic operations model. The following formula uses the information in the "percent graduating in five years" variable and in the ratio of degrees to intake:
(3)
where gradRatioPT is the ratio of the graduation rate for part-time students to that for full-time students. The "Max" and "Min" functions enter because the overall graduation rate may exceed the "% within 5 years" and the ratio of degrees to intakes may be noisy.
22. Other "Calibration" Graduation Rates: equal to the ratio of degrees (certificates in the case of non-matriculated students) to intake subject to the following floors: masters=0.9, doctoral=0.5, non-matriculated=0.9.
23. Annual Dropout Rates: the academic operations model requires annual dropout rates by student level. We compute the overall dropout rate for an intake cohort, which equals one minus the graduation rate, then use the student lifetime assumption to convert to an annual rate based on the whole student population. In other words, we apply the following formula to the data for each student level:
(4)
24. Undergraduate Applications and Yield Rates: the CyberCampus enrollment management model requires data on applications and yield rates by student segment and gender-ethnic group. These rates are determined by market reaction to the characteristics of the player-generated institution, as discussed in Technical Document 2.1 on the enrollment management model. Our database provides figures for overall applications and yield rates, but no breakdown by student segment or gender-ethnic group. (No data exist on applications and yield rates for graduate and non-matriculated students, so these will have to be determined by judgment.)
Technical Document 3.2, "Student Segmentation Analysis," describes how the market-based data for application and yield rates are derived from the game's database. Section 7 of this paper describes how the market data are processed to produce the initial PGI applications ratio and yield rate. The number of applications for the CyberCampus PGI equals the PGI applications ratio times student intake. We will have to make an assumption about whether the applications ratio for part-time undergraduates equals that for those who wish to study full-time (the market data do not differentiate), but this has not yet been incorporated into the model.
Provision of data on a select group of peer institutions will provide benchmarking information and spur player interest. Figure 7 presents our preliminary list of benchmarking variables. These variables can be calculated from the CyberCampus database. Many other potential indicators can be calculated as well, but the number of variables should be limited to avoid overwhelming the player. Suggestions for additions and/or deletions will be welcomed.
The algorithm proceeds as follows:
The game will present the player with a scrollable list of institutions, in order of decreasing prestige, with the PGI at the center of the initial display.
The student market segment analysis gives us the average number of undergraduate applications, admissions, and matriculations per institution, by student segment (stuSeg) and gender-ethnic group (geg) for each institutional segment (instSeg). See Table 4 of Td_3.2. Our task now is to transform the market data into an applications ratio and a yield rate for the player-generated institution.
The approach, described originally in Td_2.1 ("Enrollment Management"), involves blending the information for institutional segments according to how closely the PGI resembles each segment. The blending procedure makes use of so-called fuzzy logic principles, described below. First, however, we must describe how students weight the various institutional characteristics in making their applications and matriculations choices.
The student choice process will be represented by a student preference function. It may be possible to estimate this function empirically from our student demand data, but this is beyond the scope of our present project . We hope to obtain support and assistance for such a project, but as of now we must rely on a judgmentally-determined function.
Figure 8 presents a list of candidate variables for the student preference function, together with their mean values for each institutional segment and a tentative set of preference weights. (A blank in the weight column means the variable has not been included.) The data were obtained from our institutional database by extracting the average values for each institutional segment.
The applications ratio and yield rate, at the top of the figure, are highlighted because they provide calibration information for the dependent variables in the analysis. They also are candidates for the student preference function but are not currently included. The ratios were calculated by averaging the data for applications, admissions offers, and matriculations and then performing the appropriate divisions, not by averaging the ratios for the individual schools. This is consistent with the way the PGI figures will be generated.
Because separate tuition and financial aid variables are provided for public and private institutions, the weights add up to one for each institutional type. The public institution "blended tuition" figure averages in-state and out-of-state tuition using "percentage of in-state freshmen" as the blending factor. The preference function, denoted by pref[instSeg], is simply the weighted arithmetic average for the institutional segment.
The PGI's characteristics, determined the initial conditions algorithm, also can be used to compute preference function value. Let us denote this value by pref[PGI]. Then we proceed to get the PGI's applications figures and yield rates.
The first step is to compute notional figures, based on the average values in each institutional segment, for the PGI's applications, offers, and matriculations in each student segment and gender-ethnic category. Denote the figure for applications by prelimApps[stuSeg,geg]. Then:
(5)
and similarly for offers (prelimOffers) and matriculations (prelimMats). The parameter sigma () is the standard deviation of the preference values for the individual schools that make up the institutional segment. All the data for institutional segments have been computed and will be stored in the CyberCampus database. The notional values are illustrated in the first three columns of Figure 9. They will be determined as part of the game's initial conditions and then will vary during play.
Equation 5 was motivated "fuzzy logic" concepts. The preference results for institutional segments are represented by bell-shaped curves with mean equal to pref[instSeg] as shown in Figure 8 and standard deviation equal to instSeg. The current value of pref[PGI] is superimposed on the array of bell-shaped curves. Then the height of each curve is defined as winstSeg and the weighted average of appR[PGI] or yieldR[PGI] computed.
The fuzzy logic approach produces a highly non-linear yet stable result which seems ideal for CyberCampus. We should be alert to other potential applications during game development.
The second step is to calculate the PGI's overall applications ratio, shown at the bottom of the fourth column in Figure 9, based on the notional totals for applications and matriculations. The third step computes the percentage distribution of applications (also in the fourth column) by dividing each cell in the first column by the column total. The final step computes the yield rates (column 5) by dividing each row's matriculations by the associated figure for offers.
The applications ratio and yield rate procedures will be executed once per simulated year as part of the enrollment management algorithm.
Figure 2: List of Database Variables
Table continues column by column on the next page.
Figure 2: List of Database Variables (continued)
Figure 3: Geographic Region Definitions
Figure 4: Intercollegiate Athletics Base Cost and Revenue (assumes a 50-50 won-lost record)
Figure 5: Intramural athletics & Recreational Cost per Student
Figure 6: Weights for Average Salary of Adjunct Faculty (weights are applied to the indicated regular faculty salaries)
Figure 7: Peer Institution Benchmarking Variables
Figure 8: Student Preference Function Variables and Weights
Figure 9: Illustration of Application Ratios and Yield Rates for the PGI (based on an assumed set of player specifications for the PGI)
