The Ivy Institute | College Admissions Research | November 2026
Abstract
Structured advising can change things students do before they submit an application. The size and meaning of the change depend on which task is measured, what students could already do and which support the comparison group receives. This report synthesizes selected intervention studies and re-expresses published randomized estimates from two versions of a virtual advising program. In the more intensive version, assignment increased the probability of applying to two college systems by 5.4 percentage points and to any four-year college by 3.4 points. The estimate for acceptance at any four-year college was 1.7 points, with an approximate interval spanning zero. Students also reported greater access to information and support. Those reports do not establish mastery of a particular college’s requirements. External pre/post assessments also showed gains in perceived college-choice knowledge and tested scholarship-application knowledge. The second part specifies a four-week prospective study of list balance, deadline readiness, requirements knowledge and task-specific confidence. It includes parallel assessment forms, blinded review procedures and a missing-data analysis. The evidence is entirely external. In a Texas workshop, 839 paired knowledge tests rose from 0.92 to 6.62 out of 15. No effect was estimated for The Ivy Institute’s own students. Together the evidence and protocol identify which claims a family can reasonably make about progress before an admission decision exists. [R1–R4, R12–R14]
The decision a family needs to make in October
Imagine a student whose online dashboard contains twelve college names, a nearly complete general application and an essay labeled final. A parent might reasonably conclude that the family is close to submission. A short review could reveal that the intended program requires a portfolio, that a scholarship deadline falls before the general admission deadline and that the most attainable option on the list may exceed the family's budget. Each discovery changes what the student should do next. None can be detected reliably by counting completed fields in the general application. A definition of readiness must therefore describe what the student can demonstrate against the requirements of a particular route.
Now imagine a different student who has chosen three colleges, wants to remain near home and can commute to an affordable program that fits a clearly articulated goal. The shorter list might be more defensible than the list of twelve. A scale that awards points for adding schools will reward generic expansion and could actively push the second student away from a good decision. A scale that awards points for establishing a viable option, documenting uncertainty and explaining tradeoffs can distinguish the two cases. The score becomes a measure of a planning task, with no presumption that more applications or a more selective destination is better.
The word structured also needs a definition. Weekly contact can amount to four unconnected conversations. A student may feel heard without solving a single dependency. Conversely, one short meeting with a current checklist, assigned actions and a scheduled verification might produce tangible progress. In the proposed study, structure means that each session has a defined input, student action, resulting artifact and subsequent check. Information delivery alone is insufficient to satisfy that definition. This interpretation permits a meaningful fidelity measure: a reviewer can identify which parts of the process actually occurred without deciding whether the counselor was personally impressive.
There are three questions embedded in the title. Can organized assistance improve application behaviors under any conditions? Has a particular organization demonstrated improvement in its own students? Which prospective measurement could distinguish genuine improvement from the passage of time? The randomized literature addresses the first question for some interventions. The local records available for this report do not answer the second. The published instruments address the third. Keeping these levels explicit allows the report to offer a substantive empirical result while preserving the boundary around students who have not agreed to take part in research.
Evidence selection and analytic approach
This is a targeted evidence review and secondary analysis of published aggregate estimates. Searches sought original experimental studies of college application assistance, pre-college advising, virtual counseling and financial-aid completion. Additional searches covered pre/post college readiness and school-level advising. Priority was given to studies that identified an intervention, an interpretable comparator and an outcome close to the application process. Null and mixed results were deliberately retained. The search was directed toward the research question and task design rather than an exhaustive bibliographic census. A pooled effect would conceal differences in program intensity, baseline opportunity, endpoint and follow-up period.
Phillips and Reber’s V-SOURCE experiment was selected for numerical extraction because it separated two program variants and reported both intermediate and later outcomes. The numerical source is the public working paper’s Tables 5 through 8. The peer-reviewed article provides the substantive interpretation. Working-paper version, table, scale, sample size and standard errors are retained so that an editor can identify the exact numbers without assuming every edition of the paper has identical numbering. We selected outcomes directly relevant to information, assistance, breadth of application, timing or admission. The selection is not a test across every possible trial endpoint. [R2, R3]
For each binary outcome we multiplied the reported regression coefficient and its clustered standard error by one hundred to express both in percentage points. For standardized indices we kept the published standard-deviation units. As a reading aid we computed approximate 95% intervals using the reported coefficient plus or minus 1.96 times the reported standard error. Published inputs are rounded, and these intervals do not incorporate adjustments for multiple comparisons. Where the authors mark a result as surviving such an adjustment, their marking governs any claim about it. Differences between the two active arms require a covariance or a direct test. Simply subtracting two point estimates and treating their reported errors as independent would be inappropriate.
The primary student observations behind those estimates remain within the original research program. The authors discuss a consent-based administrative match. This report uses their public aggregate tables and does not obtain or republish student records. Published comparisons can be informative without trying to reverse engineer identities or recreate consent that was given to someone else. For the proposed local study, education-service agreements, participation in counseling, feedback emails and permission to share a testimonial were not treated as consent for this separate use. The local numerator and denominator remain unavailable; they are not recorded as zero improvement.
What the trials establish
In the Ontario school-based experiment by Oreopoulos and Ford, graduating students in low-transition schools were guided through a sequence of workshops. They identified programs, completed applications and worked on financial aid. The intervention increased application rates by 15 percentage points and college going by five points in the reported overall analysis. A family reading those results should distinguish a workshop in which a student actually completes a step from one in which a presenter demonstrates a step. Both may distribute similar information. Only one makes time and assistance available at the point where the student has to act. The figures refer to that study population and policy setting. [R1]
The FAFSA experiment by Bettinger and colleagues provides a related distinction between information and help carrying out a task. Assistance delivered in the tax-preparation setting made it easier for families to act on information they already had or could supply. The study reports effects on aid application and later educational outcomes, including a difference between 28% and 36% in completing two years of college within three years for the relevant high school senior group. Those percentages describe a specific downstream outcome. They cannot be repackaged as a measured change in deadline readiness or as a general return to hiring a counselor. The experiment motivates observation of a student's completed work product in addition to asking whether the family has heard the instructions. [R4]
A New Hampshire mentoring study by Carrell and Sacerdote found a substantial college-going effect among women assigned to treatment and much smaller effects among men. The authors discuss differences in alternatives and circumstances. That finding offers a warning about incentive and opportunity outside the advising session. A student can understand an application perfectly and still make a rational choice to pursue work or delay college. A local evaluator should not code that choice as a failure of comprehension. The readiness task asks whether the student can make and execute an informed plan; deciding against a particular institution can be a successful demonstration of that capacity. [R5]
At larger scale, Bird and colleagues examined information campaigns concerning financial aid that collectively reached more than 800,000 students. The results caution against assuming that reminders successful in a local setting will necessarily change behavior in a very large campaign. The contrast with hands-on workshops concerns the tasks a student must complete and the help available when a barrier appears. It would be speculative to attribute every difference between these studies to human contact alone. Settings, populations and implementation also varied. Still, when a local program calls itself structured, it is reasonable to document whether it actually provided an opportunity to resolve a blocked task. [R6]
A randomized evaluation of a near-peer counselor placed in schools found no significant overall enrollment effect pooled across its first three years, alongside evidence of gains in some targeted student groups. The unit of assignment matters. A whole school can benefit from changes in counselor routines or information sharing even when an individual student never reports meeting the assigned adviser. The program effect answers a question about offering a resource in a school. It is not a causal estimate of the benefit of an additional private hour for a student who deliberately sought it. Such distinctions are particularly important when families compare evaluations from public-access programs with commercial advising claims. [R7]
Barr and Castleman examine intensive support extending across high school and college. Their randomized evidence connects advising with bachelor’s degree attainment among lower-income students, with initial enrollment quality important to the interpretation. That study asks a longer-horizon question than the present protocol. Ongoing help after enrollment could change persistence even if two students had equally complete application plans in high school. For families, that distinction is substantive: choosing a program that provides continuing support is a different purchase or enrollment decision from choosing help limited to the application season. For researchers it determines when a baseline should occur and which intervention is under study. [R8]
A small randomized pilot deposited in Harvard’s research repository provides an especially relevant counterpoint. It followed 107 high-achieving seniors from relatively low-income families and randomly offered 52 of them ten hours of college counseling. The authors found little or no effect on application quality, while the choice of colleges appeared to change. Their estimated difference in enrollment at the most competitive institutions was not statistically significant. A local report should not cite that estimate as proof that counseling improved admission chances. It also should not take a null application-quality result to mean that all possible kinds of preparation were unchanged. An artifact rubric must specify the construct it aims to capture. [R9]
Clayton and Worsham provide a pre/post study at the school level using comparison schools in North Carolina. The introduction of an adviser was associated with more students taking the SAT and a lower school average score. One plausible interpretation of that combination is a change in who took the test. A decline in a group mean does not by itself imply that individual students lost skill. The authors’ unit of analysis and outcome differ from ours, yet the lesson is directly applicable: a program that reaches more previously underserved students can alter the observed composition of a scored group. Report participation and missingness alongside any before-and-after mean. [R11]
What changed in a virtual advising experiment?
| Outcome (unit) | Milestones effect [95% interval] | Complete effect [95% interval] | N |
|---|---|---|---|
| Had information (SD) | 0.086 [0.035, 0.137] | 0.109 [0.050, 0.168] | 5,993 |
| Had support (SD) | 0.080 [0.029, 0.131] | 0.152 [0.099, 0.205] | 5,931 |
| Sought information (SD) | -0.033 [-0.086, 0.020] | 0.017 [-0.044, 0.078] | 5,986 |
| Applied to two systems (pp) | 0.6 [-1.9, 3.1] | 5.4 [2.9, 7.9] | 5,986 |
| Filed aid application on time (pp) | 2.8 [0.6, 5.0] | 1.7 [-0.7, 4.1] | 6,640 |
| Applied to any four-year college (pp) | 2.5 [0.5, 4.5] | 3.4 [1.2, 5.6] | 5,986 |
| Applied to any selective college (pp) | -0.0 [-2.4, 2.4] | 3.6 [1.1, 6.1] | 5,986 |
| Accepted by any four-year college (pp) | 0.0 [-2.2, 2.2] | 1.7 [-0.8, 4.2] | 5,986 |
SD = control-group standard deviations. pp = percentage points. The comparison is each treatment assignment versus control. Intervals use rounded, school-clustered standard errors. Source [R3], Tables 5–8.
The table provides a useful ordering of evidence. Both variants increased students’ sense that they had information and support. The more intensive variant produced a larger point estimate for reported support. A student can feel supported because someone responds to a message, explains an unfamiliar phrase or takes an interest in a decision. None of those requires that a student be able to verify a complex requirement unaided. Conversely, a student who independently manages a difficult requirement may feel dissatisfied with the help received. The study’s scales are valuable on their own terms. They should not be relabeled as observed competence or as a direct test of confidence in the proposed local tasks.
The contrast between applying and receiving an offer is visible in the same set of students. Assignment to the Complete variant increased the probability of applying to both systems by 5.4 percentage points, against a reported control mean of 48.9%. That adjusted coefficient implies approximately 5.4 additional such applications per hundred assigned students under that comparison, counting students rather than counting extra individual forms. It does not show that the new applications formed a better-balanced list. Two systems can contain similarly selective or similarly unaffordable options. A reliable local instrument needs to inspect the options and constraints behind the number.
For applications to any four-year college, the Complete coefficient was 3.4 points and the reported standard error was 1.1 points. A conventional approximate interval is 1.2 to 5.6 points after rounding. The acceptance coefficient for any four-year college was 1.7 points with a standard error of 1.3 points, giving approximately −0.8 to 4.2 points. That interval includes modest positive values. It also includes zero. It is appropriate to say that the study did not establish an acceptance gain on this endpoint. It would be too strong to say it proved the effect was exactly zero. [R3]
The timing endpoint deserves its own reading. Filing the FAFSA on time was measured with state administrative data, while the two-systems application measure was self-reported. Those evidence types have distinct strengths. Administrative evidence can establish that a transaction reached a particular system in time. Self-report can capture activity outside a single administrative source but may introduce recall or reporting differences. The Complete estimate for on-time aid filing, 1.7 percentage points, was imprecise in this comparison. The authors’ table also identifies which outcomes survive adjustment for testing multiple hypotheses. Whether a confidence interval crosses zero is only one part of that interpretation.
These comparisons describe assignment to services within an experiment, not an estimate obtained by selectively comparing students who attended the most meetings with students who attended none. The latter comparison would also reflect differences in motivation, scheduling flexibility, underlying need and ability to reach the program. The local protocol therefore logs attendance and work completed but does not turn a high-attendance subgroup into a causal estimate. If a program director wants to know what would happen if participation increased, the design would need an appropriate source of variation in participation. A correlation between attendance and progress cannot supply it on its own.
External before-and-after evidence across cohorts and programs
The Washington State Institute for Public Policy evaluated Make It Happen!, a four-day residential college program for young people in foster care. Workshops, a college fair and encounters with college students and faculty formed a defined experience. The public appendix prints pre- and post-program percentages separately for high-school and college-bound groups in 2005, 2006 and 2007. This makes it possible to answer a before-and-after question now using external observations. It does not require access to any current client’s application. We transcribed the displayed percentages, calculated post minus pre in percentage points, and retained the distinction between younger participants and those already preparing to enter college. [R12]
For the high-school group, the average proportion of affirmative answers to the application-knowledge items rose from 42% to 93% in 2005, from 41% to 82% in 2006, and from 52% to 87% in 2007. Those figures describe the program’s getting-to-college composite, an average across item percentages. They do not mean that 93% of individual students satisfied a readiness standard. Questions changed across years. The proper comparison is within a year using that year’s items. The three gains range from 35 to 51 percentage points. Collapsing them into a single pooled student effect would require information about individual responses, denominators and item equivalence that the aggregate exhibits do not provide.
High-school group. Bars show average affirmative responses to the annual getting-to-college item set. Items differ by year. Source [R12], Exhibits A.1–A.3.
| Cohort | Choice of college | Requirements | Completing aid form |
|---|---|---|---|
| 2005 | 50% → 97% | 45% → 95% | 25% → 95% |
| 2006 | 56% → 89% | 41% → 86% | 26% → 93% |
| 2007 | 47% → 89% | 52% → 86% | 23% → 77% |
Choosing an appropriate college was one of the closest available external indicators to the proposed list-balance domain. Among high-school participants, endorsement rose from 50% to 97% in 2005, 56% to 89% in 2006, and 47% to 89% in 2007. The item wording shifted slightly. A student agreeing that they know how to choose a college has expressed greater perceived capability. They have not demonstrated that their actual list contains an attainable and financially viable choice. The distinction is empirically important. A public program can show a large and repeated gain on the first construct while supplying no direct observation of the second. [R12]
The college-bound comparison within the same report is equally instructive. In 2005 its getting-to-college composite went from 89% to 84%, while the high-school group gained 51 points. The former group began close to the upper limit. They were also at a different transition stage and may have attended for different reasons. They are not a control group for the younger students. By 2007 the college-bound composite rose from 80% to 94%. Reporting the negative as well as the positive changes reduces the temptation to select only supportive results. The tables do not identify each person’s change or the standard deviation of the paired differences. Confidence intervals for individual change cannot be reconstructed from these marginals.
One could ask whether students simply became more enthusiastic about all things related to college. The report contains separate career and academic expectation items. For example, in the 2007 high-school group the proportion expecting to attend college was 89% before and 88% afterward, even as the college-choice item rose substantially. A difference between measures in the same program does not prove the absence of response bias, but it does make the distinction between learning about the process and forming an aspiration visible. Students who already plan to attend can learn how to act on that plan without changing whether they expect to attend. [R12]
A second external source provides a genuine paired learning measure. In Murr’s Texas scholarship-workshop evaluation, 839 attendees completed the paired general-learning assessment. The six questions together could yield fifteen points. The mean was 0.92 beforehand and 6.62 afterward, an increase of 5.70 points on the published scale. Using the reported standard error of the paired difference, 0.111, gives an approximate interval from 5.48 to 5.92 points. The estimate describes the observed paired participants. Test completion, repeated exposure to material and the absence of an untreated comparison constrain attribution. Participants still averaged substantially less than the maximum after the workshop. A large gain from a low baseline and incomplete mastery can coexist. [R13]
This example also shows why primary tables should be checked for internal consistency. In the same paper, the separate essay-paragraph analysis reports a test statistic and p value that do not agree under an ordinary paired t test with the listed degrees of freedom. The methods describe a one-to-ten rating, yet the stated pretest mean for the essay endpoint is below one. We do not use that endpoint to support or refute an effect. The general-learning endpoint has a different scale, denominator and internally consistent order of magnitude. Excluding one inconsistent endpoint is a narrower judgment than excluding the entire evaluation. It also avoids mixing the headline participant count with the actual number contributing to a specific result. [R13]
A third example tests whether external reviewers can score redacted work. Early and DeCosta-Smith studied instruction in the college admission essay in two senior English classes. Two former teachers scored anonymized, mixed-order work independently. The authors reported a rank correlation of 0.76 between ratings. In their essay table the instructional group moved from 3.97 to 4.91, while the comparison group moved from 4.25 to 3.68. The table labels the means adjusted and prints 17 and 22 students respectively. The self-efficacy table prints 16 and 21. Its instructional-group means moved from 5.91 to 8.06 compared with 5.85 to 5.69 in the other group. Both the artifact and the self-report are real pre/post observations. Neither represents a measurement of the proposed all-purpose readiness score. [R14]
The essay study assigned two intact classes by coin toss. With one class per condition, class composition, meeting time and other classroom conditions cannot be distinguished cleanly from treatment at the cluster level. The instructional group’s higher posttest mean can be reported as observed. It is not a general estimate of what would happen in every counseling setting. Likewise, correlated reviewer scores indicate that reviewers tend to order essays similarly. They do not establish perfect agreement at the same score level. These design details inform the proposed use of reviewers across multiple counselors and the decision to report absolute agreement together with a correlation-based measure when appropriate.
Taken together, these public data directly document changes in perceived college-choice ability, scholarship-procedure knowledge and independently assessed writing. A separate virtual trial supplies an administrative timeliness endpoint and comparison-group estimates for application breadth. The sources leave a specific gap. None of the reviewed cohorts contains the same students’ verified financial and academic list balance, a dated dependency plan, a parallel requirements test, and a separate confidence score measured together. The present report therefore supplies domain-specific findings without imputing the missing joint distribution. The prospective protocol below would fill that gap for any participating organization.
Data discovery register and boundaries of coverage
To extend the search beyond positive program reports, the review checked journal articles and their working-paper versions, public evaluation reports, research repositories, author project pages, the federal education evidence clearinghouse and a public replication catalog. Discovery concepts combined college or scholarship application, advising or counseling, multiyear panel or before/after, rubric, confidence, placement and randomized assignment. Duplicate versions were grouped by study. A peer-reviewed article and the table-bearing working paper from the same trial count once. An institutional press release covering a trial is also not an independent replication.
The V-SOURCE authors’ replication catalog provides a route to their published project files. The current reanalysis uses the tables and their supplied errors, rather than representing an aggregate extraction as a participant-level model rerun. NCES’s High School Longitudinal Study is an additional route for studying nationally observed associations among advising, student plans and transitions. Its existence does not establish measurement of these exact four domains at matching time points under a defined treatment. Using a broad national dataset without mapping the questionnaire items and selection mechanism would add sample size but weaken the meaning of the treatment. The data-source register includes both routes so a subsequent researcher can test that mapping explicitly. [R15, R16]
A UCLA research poster on rising ninth graders in My Journey to College used a pretest/posttest design and reported gains in college knowledge, with no significant change in attitudes toward the college-going process. The participants were younger than the local cohort proposed here, and the poster is a limited account of the study. It is evidence that knowledge and attitude can behave differently even in a college-preparation setting. It is not a replication among seniors. Similarly, published research on parents’ college-preparedness knowledge provides a different respondent. A change in what parents know may affect household planning, but a parent response should never be substituted for an absent student response in a paired student analysis. [R17, R18]
Other accessible materials include a college-knowledge outreach camp case study and a USC Pullias Center description of collecting data before and after a digital college-access intervention. These sources are useful for studying implementation and the logistics of measurement. A description of why data are being collected is not itself an effect estimate. Our evidence register indicates where a source provides outcomes, where it describes methods and where it simply identifies potential data for future work. Publicly accessible documentation makes a study inspectable. It cannot by itself solve the problem of selecting an endpoint appropriate to a family’s decision. [R19, R20]
Translating evidence into a measurable local question
The proposed primary question is how much artifact-based application readiness changes from just before a defined four-week sequence to just after it, among participating first-year undergraduate applicants who provide appropriate research permission. The interpretation is an observed paired change during the advising period. A separate question concerns attribution to advising. Addressing it would require a credible comparison condition, such as a randomized sequence of invitations where capacity and ethical review permit, or a concurrent group with comparable deadlines, prior preparation and usual support. The single-group protocol does not automatically acquire a causal interpretation because it is prospective.
The primary outcome is the sum of three equally weighted domains: defensibility of the college list, readiness for deadlines, and demonstrated understanding of requirements. Each has three indicators scored zero to three, making a total from zero to twenty-seven. Confidence uses a separate four-item student measure. It is kept separate because feeling certain about an incorrect plan should not increase an objective readiness score. The aggregate score is a design proposal, not a published validated scale. Domain-level results should always be shown. A gain in list quality cannot compensate, for practical purposes, for a missed immutable deadline at the student’s top viable choice.
Eligibility for the proposed cohort is defined by application stage. A participant must be working toward a first-year undergraduate application, have at least four weeks in which to complete the sequence before the earliest relevant external deadline or be able to undertake an equivalent planning exercise, and agree to the relevant research procedures. Record which route was used. A participant with rolling deadlines and a participant whose only program closes next week face different work. Students may receive ordinary support regardless of research eligibility. Students who decline use of their work should still be able to use the same advising materials in their regular service.
The preferred baseline occurs before the first structured session. The student completes an artifact submission and brief assessment independently, using the allowed sources and accessibility supports. Record whether the student has already had comparable support elsewhere. The exit assessment occurs within seven days of the final session, using the parallel form. An optional later assessment about one month after exit checks whether the student can maintain a plan when a deadline or program changes. Report any work done after assessment windows in a deviation log. Time ordering matters because a baseline taken after a counselor fixes the list will understate the observable change during the intervention.
Target an initial feasibility cohort of approximately forty enrolled volunteers, without pretending that this number guarantees a particular detectable effect. With thirty paired observations and an assumed standard deviation of paired differences of four rubric points, the standard error of the mean change would be about 0.73 points. A two-sided t interval would have a half-width of about 1.5 points. That is an illustration of precision under a hypothetical spread, not a power calculation based on observed local data. Before a larger confirmatory study, replace the assumed variance with a justified estimate and address clustering by counselor and school.
Four sessions with an observable product
Session one: build a defensible set of options
Ask the student to state what they most need from a college, including academic access, location, affordability and nonnegotiable constraints they choose to disclose. The program records affordability as a student-defined category or broad feasibility status. It does not need tax returns or private household figures for research. Together student and counselor inspect a preliminary list, testing whether any currently plausible option meets those conditions. The counselor explains how uncertainty in a program’s selectivity or aid can alter the balance. The student then updates the list and gives a reason for each substantive change.
The session ends with a tangible work product. For each option the student records a program, an evidence source, a reason to include it and at least one condition that could change the judgment. If no affordable, academically viable option can be documented yet, the student records the gap and a dated verification plan. This is more honest than labeling a highly speculative scholarship a financial safety. A research reviewer scores the evidence of planning and resolution rather than the price of the colleges selected. Recording a genuine unresolved constraint can earn partial credit while protecting a family from false reassurance.
Session two: turn dates into responsibilities
The counselor and student trace one application backward from its earliest consequence-bearing deadline. An early scholarship deadline may govern the timeline even if the general application stays open. A portfolio or audition may require registration before its materials are due. A counselor form may be outside the student's control. The student identifies the source, the date, who will act, what must precede the action and how receipt can be verified. Record local time and time zone when the institution provides them. When it does not, record the ambiguity and a conservative internal target without presenting it as an official deadline.
The work product is a calendar connected to a dependency list. Each row represents a specific deliverable and owner. Students can choose an accessible format: a spreadsheet, paper planner or electronic reminder system can all earn the same score if the content is equivalent. The research export standardizes the fields so a grader cannot reward a polished template over a student’s plain table. A missed official deadline is documented accurately. Backdating an internal target to make the calendar look orderly is an error. A workable alternative route or documented contact can still demonstrate recovery planning.
Session three: verify requirements and teach them back
Begin with a plausible source conflict. A generic admissions page may describe a standard route while a major-specific page adds a requirement. The student follows the institutional navigation to identify the applicable cycle and program, notes the discrepancy, and decides whether an official inquiry is necessary. The counselor asks the student to explain what would happen if the conflicting requirement were ignored. This teaches the student to locate the controlling source. The scoring exercise uses a fictional policy packet so that the final test measures applying a rule and does not reward memorizing the exact college discussed in the session.
The resulting evidence is a requirements matrix and a short explanation in the student’s own words. An accurate note that a requirement is conditional can be stronger than an extensive list of requirements copied from unrelated programs. A grader checks source, applicability, interpretation and proposed action. Students who use speech-to-text, translation support for accessing material or approved extra time should have those accommodations documented consistently across occasions. The study measures navigation and decision making rather than composition style. The student may ask for clarification of an instruction but must make the scored decision independently.
Session four: rehearse submission and recovery
Present a late change that forces the student to update the plan. A recommender might be unavailable; a portfolio site might require its own registration; the preferred route might become financially infeasible. The student rehearses the next action without submitting a real application, paying a fee or asking a teacher to perform work solely for research. The counselor asks where evidence of receipt would appear, what the backup is and when it would be appropriate to seek help. Students should not be rewarded for claiming they control another person’s response.
At the end, the student prepares an independent exit packet and confidence ratings. The counselor completes a separate fidelity checklist covering session dates, activities completed, any material adaptations and supports provided by someone else. The instructor does not grade their own student. The export made available to reviewers removes names, recognizable personal stories, institution-specific identifying details where feasible, dates that reveal wave and indicators of who provided counseling. If an essential detail cannot be redacted without making a decision unscoreable, substitute a standardized task for both occasions or mark the indicator missing with a reason.
Published scoring manual for redacted work
Each indicator is scored zero to three. The zero anchor reflects observed absence or an incorrect answer in a packet the student submitted. A missing packet receives a missing code and an explanation where available. Mixing the two would make nonparticipation look like a complete lack of preparation. Scores of one, two and three represent increasingly usable evidence. A student can earn full credit with brief, plain answers. Elaborate prose, expensive activities and a recognizable school name confer no scoring advantage. The examples in this manual are constructed for calibration and contain no student records.
L1. Academic and personal fit
Award zero when the list contains names without an identifiable academic or personal rationale, or when the stated program does not exist in the specified route and no uncertainty is acknowledged. Award one for a general preference attached to a plausible option. Award two when the student connects an actual offering or access condition to an articulated priority. Award three when that connection is supported by a relevant source and the student identifies a material tradeoff between plausible options. The tradeoff can involve teaching format, commute, access to a major, flexibility to explore or another criterion the student chooses.
For example, a sentence stating that a college is excellent for engineering gives little evidence about what the student has checked. Explaining that students must apply separately to enter the intended major, and why that matters if a change of major is difficult, shows a more specific understanding. A student undecided about a major can score equally well by explaining which exploratory routes remain open. The rubric should not require premature specialization. The research question is whether the list reflects an informed decision under the student’s own circumstances.
L2. Plausible academic routes under uncertainty
Award zero for an unsupported assertion that every college will admit the student or for the absence of any consideration of realistic access. Award one for labels such as reach or likely without an explanation. Award two for a reasoned assessment of variation in access, including at least one plausible enrollment pathway or a clearly identified gap. Award three for evidence relevant to the actual program or route plus an actionable alternative if the most uncertain options fail. A documented transfer or open-access pathway can qualify when it fits the student’s goal.
The reviewer must not supply an admissions probability. A published overall admission rate does not account for a particular student, major or selection system. A student who correctly recognizes that a source is too coarse and seeks a more relevant figure should receive credit for reasoning, not a penalty for declining false precision. Conversely, a counselor’s undocumented classification is not enough by itself. The student should be able to explain which assumption matters and how a change in that assumption would affect the list.
L3. Financial feasibility and an alternative
Award zero for ignoring a stated affordability constraint. Award one for labeling an option affordable without indicating the basis. Award two for identifying an appropriate cost estimate or funding source and acknowledging uncertainty. Award three for identifying a feasible route under a conservative scenario, or documenting a genuinely unresolved barrier together with a specific, timely alternative that could protect access. A student does not need to disclose family income to the reviewer. A redacted record can say that the estimated net price exceeds the family’s stated range.
An advertised maximum scholarship does not establish the amount a particular student will receive. The reviewer should look for whether the student distinguishes a confirmed award, a conditional estimate and a competitive possibility. If a local commuting option fits the academic plan and household constraints, it can be strong evidence of balance. If even the least expensive known route is outside the family’s range, the reviewer should record that the affordability flag is unresolved regardless of total score. The measure evaluates planning; it cannot create financial resources for a family.
D1. Official dates and applicable cycles
Award zero where the consequential due date is absent or clearly from an inapplicable cycle. Award one for recording a date without a source, cycle or match to the selected route. Award two for the correct institutional date linked to the appropriate route. Award three when the plan also distinguishes among relevant institutional deadlines, such as general admission, priority scholarship and program-specific receipt dates, and records when the information was checked. If a future cycle is not posted, a carefully labeled provisional date with a verification trigger can receive partial credit.
This indicator tests what the institution actually requires. An internal deadline can be excellent planning but should not be represented as the official due date. A student and counselor who disagree about an ambiguous date can retain both interpretations, identify the office to ask and choose an earlier working target until the answer arrives. Reviewers should judge that as managed uncertainty. They should not reward a confidently stated date simply because it looks cleaner on a calendar. Accuracy and confidence are deliberately recorded on different scales.
D2. Dependencies and ownership
Award zero when essential steps lack an owner or cannot logically be completed in the sequence shown. Award one for a generic to-do list without responsibility or lead time. Award two when major dependencies and owners are identified. Award three when the plan works backward from consequential dates, assigns a feasible internal target and states what the student will do if an outside person or system is delayed. The student must distinguish asking someone for a document from guaranteeing that the document will arrive.
A teacher recommendation illustrates the problem. A calendar entry saying submit recommendation next Friday may assign an action the student cannot perform. A stronger plan identifies who requests the recommendation, who uploads it, how much notice is reasonable under the school’s own process and where the student can check status. The measure does not require providing a teacher’s name. A role label is sufficient. Slow response by a third party is not scored against the student if the student has correctly recognized and managed the dependency.
D3. Verification and recovery
Award zero if a plan treats an unchecked draft as a verified submission or overlooks an essential missing item. Award one for a broad intention to check later. Award two when the student identifies the appropriate receipt, portal status or confirmation for a consequential item and a specific time to check it. Award three when an appropriate recovery route is also specified, with an escalation threshold and attention to whether duplicate submissions would cause problems. The rehearsal may use fictional status messages; no live application is sent for testing.
A family may have evidence that a fee was processed without evidence that a portfolio uploaded successfully. The student should identify what the receipt confirms and what remains unknown. An email draft to a designated office can demonstrate a recovery plan, but its wording should avoid disclosing unrelated personal information. Reviewers should give equal credit for an equivalent phone or accessible support route. They should examine whether the plan could work before the consequence occurs and whether the student knows what to preserve as proof.
K1. Find the controlling source
Award zero for relying on an inapplicable source while ignoring authoritative information available in the packet. Award one for naming a plausible source without locating the relevant rule. Award two for selecting the correct institution, program and cycle. Award three for explaining why a competing piece of information applies to a different route, is superseded or requires further confirmation. A source need not have the most recent date to apply; a current supplemental page can incorporate an older standing requirement.
The parallel assessment supplies a short general handbook, a program addendum and an unofficial summary. It asks the student to decide which provisions govern a particular applicant. A test of browsing speed would disadvantage students with unreliable internet, unfamiliar screen-reader behavior or less practice with a particular website. Supplying equivalent source packets keeps the comparison focused on interpretation. A separate transfer task can invite the student to verify one item on a real public website and record whether technical barriers were encountered.
K2. Apply a conditional rule to a new case
Award zero if the answer reverses a condition or assigns a requirement to the wrong applicant. Award one for naming the relevant requirement without correctly applying its condition. Award two for a correct decision and explanation in the case presented. Award three when the student can also explain how the decision changes in a specified counterexample. The counterexample should alter one relevant condition at a time. Adding multiple differences makes it impossible to tell which rule the student learned.
Suppose an additional sample is required for direct entry to a studio program and optional for general admission. A student pursuing direct entry should identify the additional sample as required for that route. The student choosing general admission could correctly call it optional while still investigating whether that route permits later entry to the studio. These decisions can be equally accurate. A scoring rule that always rewards collecting more materials would encourage unnecessary work and could hide misunderstanding of the program-specific consequence.
K3. Explain an unresolved issue and the next step
Award zero if the student invents a rule or treats silence as proof that an important restriction does not exist. Award one for an unspecific statement that more research is needed. Award two for identifying the unanswered question and the appropriate issuing office or document. Award three for a concise inquiry, a conservative interim plan and a description of how the answer would alter the application. Students may use a template or dictate the proposed inquiry. Tone and writing elegance are not part of the score.
Some well-prepared students become less confident when they learn how many rules depend on route and year. A reviewer who rewards only certainty will misclassify intellectual caution as a deficit. This indicator rewards the capacity to proceed responsibly despite an information gap. To prevent blanket uncertainty from becoming an easy way to receive full marks, the student must identify precisely what the available evidence settles and what it leaves unresolved. An answer that asks the university to repeat a clear published rule does not demonstrate this skill.
Student confidence and correspondence with performance
Ask the student to rate four capabilities from zero to four immediately before seeing the scored task at each occasion. Use identical anchors at baseline and exit. Zero means I could not start without someone doing this with me. One means I could start with step-by-step help. Two means I could finish with a check from someone else. Three means I could finish independently using a source. Four means I could do it independently and explain how to recover if the information changed. Offer a separate prefer-not-to-answer option. It is a missing confidence response and does not change the artifact score.
The four capabilities are explaining why my list contains a feasible option, building a realistic schedule from an official deadline, deciding whether a program-specific requirement applies, and identifying what to do if submitted material is marked missing. Report each item, then the mean across all four when all are present. If one is missing, report available item responses but leave the prespecified complete confidence index missing. Explain that the scale is local and task-specific. It has not been validated as a measure of general self-efficacy, anxiety or motivation.
Interpret confidence alongside demonstrated work. A student may enter with high certainty and a mistaken reading of a program rule, then leave with moderate certainty and an accurate contingency plan. The artifact improved even though confidence fell. Another may enter with low certainty and already accurate work, then leave with unchanged work and greater willingness to proceed independently. Those trajectories answer different questions. For group reporting, show cross-tabulated directions of change where cell sizes permit, without implying a common numerical scale. Subtracting a four-point confidence score from a twenty-seven-point preparation score would produce an uninterpretable number.
Positive self-report can reflect encouragement, relief or familiarity with a counselor in addition to mastery. The exit survey should be returned to someone outside the counseling relationship where feasible, with an explicit statement that honest responses do not affect service. Asking the student to please show us how much the program helped would prime a particular response and confound satisfaction with capability. At a delayed assessment, ask the same four questions before a new practice case. A student who remains confident after encountering an unfamiliar rule provides different evidence from one who remembers how the instructor solved the exact prior problem.
Parallel assessment materials
Form A: a fall studio applicant
The following institutions and dates are fictional teaching materials. Alder College’s general guide says that first-year applications close December 10 and supporting school documents may arrive by December 15. The studio addendum states that direct-entry studio candidates must register a portfolio account by November 20 and submit work by December 1. The access-award page says that students seeking consideration must submit the general application by November 25. Award decisions are competitive. A third-party guide mentions a December 20 application deadline without naming its cycle. A student wants direct studio entry, is seeking the award and needs an affordable alternative.
Ask the student for the controlling dates and their order; a short plan for the school document; an explanation of why the third-party date is insufficient; the financial uncertainty that remains; and one action if the portfolio dashboard still shows pending after an upload. Supply a second fictional school with a commuting option, an estimated cost within the student’s range and studio access conditional on a first-year review. The student may include or exclude it if the reasoning explains the tradeoff. The exercise measures whether the student notices access and cost, not whether they choose the option preferred by the author.
The answer key recognizes November 20 as the portfolio-registration step, November 25 as the general application target for award consideration, December 1 as the portfolio deadline and December 15 as the separate school-document deadline. December 10 remains the regular general deadline. A student who applies December 10 could be on time for admission while missing the scholarship consideration date. The student should avoid promising a scholarship, distinguish formal receipt from uploading and identify the correct office or status tool. A technically accurate timeline with no route the student could reasonably afford remains financially unresolved.
Form B: a spring design applicant
Birch University’s general guide says that applications close February 12 and school documents may arrive by February 18. The design pathway requires students to reserve a review slot by January 19 and submit a work sample by February 3. An access-award notice says that applicants must file the general form by January 26 to enter consideration. The award is competitive. A discussion-board post describes a February 21 deadline without identifying its year or pathway. A student seeks the design route and the award. An alternative commuting pathway appears financially feasible, but later admission to design depends on a separate review.
Ask for the same five products as in Form A, varying surface details to discourage memorization. The intended sequence is January 19, January 26, February 3, February 12 and February 18, with the conditions attached to each. A student who gives February 12 for every job misunderstands the difference between formal general admission and a viable application for the intended pathway. A student who proposes January 15 as an internal working date can earn credit if they clearly identify it as their own margin for error. A cautious target does not become the university’s published requirement.
Half of participants should receive A before B and the remainder B before A, using a predefined order assignment. In a small feasibility cohort record exact counts in each order. If one form consistently yields lower scores even among comparable baseline students, report the imbalance and consider revision before interpreting the magnitude of change. Do not relabel the easier task as the exit instrument after seeing the data. The packet can rotate invented names and calendar months without changing the number of conditions or dependencies. Any substantive change deserves a new instrument version and a record of which participants used it.
Independent scoring, consent and governance
Two reviewers with experience reading application plans should score each redacted artifact separately. Neither should have advised the student or be compensated according to the direction of the study result. Randomize packet order, assign opaque identifiers and remove labels revealing which assessment came first. Store the crosswalk with a steward who does not score the work. The occasional content clue may still reveal an occasion, for example an explicit sentence about a recent workshop. Record suspected unblinding and repeat the main summary without those packets as a sensitivity check if enough observations remain.
Before production scoring, the reviewers independently rate calibration materials and discuss the reason for disagreement indicator by indicator. Preserve the original ratings. If the manual changes, assign a new version and re-score all study packets under the same rules. For the actual analysis retain each original score, report exact and within-one agreement by indicator and identify where ambiguities concentrate. An ordinal weighted kappa can supplement the simple agreement measures where there is adequate variation. For the total score, report the chosen absolute-agreement intraclass correlation model and its interval if estimable. A correlation that ignores systematic differences between raters is insufficient.
For the prespecified final participant score, use the mean of the two reviewers’ indicator ratings. It may contain halves. If an indicator differs by at least two points, a third reviewer independently assesses the packet and documents the reason. Retain the original mean for the primary analysis and show an adjudicated sensitivity score using the median of the three on flagged indicators. This rule avoids silently replacing inconvenient judgments and supports a practical quality-improvement conversation. If an entire domain turns out to be too ambiguous to score reliably, report the problem rather than treating the combined total as a precise measure.
The invitation should explain what the research measures, what tasks it adds, who will see the data, how long they will be retained and how participation can end. Obtain an appropriate determination about ethics review and applicable rules before enrolling anyone. Where parental permission and student assent are required, obtain both, and seek the student's own informed agreement even when a guardian is enthusiastic. Adult participants consent for themselves as applicable. The HHS guidance covers research with children under its framework; whether that framework applies to a particular private activity requires a qualified determination. The study has no local approval or enrollment to report. [R10]
A family must be able to decline without losing ordinary advising, changing fees or affecting a counselor’s support. Use a recruitment contact outside the direct service relationship where possible. For a student who withdraws, explain which identifiable data can still be removed and when data already aggregated or irreversibly de-identified can no longer be retrieved. Do not promise a particular withdrawal deadline until the steward and review oversight have approved one. Any public quotation, identifiable narrative or use in marketing would require separate consideration and permission. The proposed analysis can be performed entirely with numerical scores and brief nonidentifying evidence notes.
A practical separation of files reduces risk. The steward holds the consent record and participant key in a restricted location. An analyst receives coded rows and occasion labels only after the reviewers finish. A public release contains aggregate evidence with small cells suppressed when identification is plausible. Collecting exact family income, disability details, school name or an unredacted admissions essay is unnecessary for the standard practice cases. Redaction should remove metadata and combinations of details as well as names. Blank data templates are provided for future approved use; no student information was entered to prepare this publication.
Statistical analysis and interpretation
For student i calculate the exit minus baseline difference in the complete 27-point score. The primary descriptive result is the unweighted mean paired difference among students with all primary indicators available at both assessments. Also report the median, interquartile range, range of observed changes and counts improving, unchanged or declining. Display domains separately. A primary paired analysis weights each participant once. Averaging all completed tasks across all students instead would give greater influence to students with more complete records and could exaggerate improvement if later tasks are easier or more frequently submitted.
An approximate t interval for the mean paired difference uses the sample standard deviation of those differences divided by the square root of the number of pairs. If the distribution is strongly skewed, show the empirical distribution and a participant-level bootstrap sensitivity interval where the sample supports it. For binary milestones, such as having at least one feasible and affordable route, report the four transition counts. Students who moved from no to yes and those who moved from yes to no carry the information about change. A methods paragraph claiming a paired test without a linked measure for each student would be misleading.
The binary usable-list milestone requires an explicit definition in advance: both a plausible entry route and a credible financial pathway supported by evidence, with neither left wholly unresolved. It is a practical reporting threshold. The committee defining it should test examples from students with different educational plans, including community college, commuting and direct-entry professional programs. Report the resulting count out of complete pairs, alongside the quantitative list domain. Dichotomizing a detailed score sacrifices information. Its value is that a family can understand the practical state a student reached, not that a single cutoff is scientifically privileged.
Account for missingness at three levels. A student may decline research altogether, submit only one wave, or return a wave with a single incomplete indicator. Record each separately, with reasons only when voluntarily supplied. Show enrolled participants, baseline submissions, exit submissions, matched pairs, missing indicators and withdrawals. A returning artifact that demonstrates no viable plan receives its earned score. An unreturned artifact remains unknown. If participants who struggled were least likely to return, a complete-case estimate can be optimistic even when every individual score was correctly calculated.
For a transparent sensitivity bound when baseline data exist, replace each missing exit total with zero, then with twenty-seven, and recompute the mean difference over the full baseline cohort. These will often be very wide bounds, especially in a small study. Narrower scenarios can assume that missing students had no improvement or had changes similar to the lowest observed quartile, clearly identified as assumptions. If a claimed improvement disappears under modest assumptions, the report should say so. Never replace missing outcomes by attendance or by a counselor’s retrospective impression that the student probably improved.
Separate the practical meaning of a change from its statistical precision. Adding two score points by locating an optional brochure does not have the same operational value as resolving an unrecognized portfolio deadline. The rubric already attempts to weight essential reasoning, but any composite will compress details. Display changes in unresolved critical flags, particularly the earliest consequential deadline and the existence of a financially feasible route. A student can increase the total and still require urgent assistance. The score is a research instrument. It does not decide access to services, classify applicants for colleges or set fees.
For a causal extension, if capacity is naturally constrained and oversight approves, randomly assign the timing of an offer within an interval that keeps all students away from impending deadlines. Everyone continues to receive usual support. Compare groups at a common time before the delayed group begins. Retain all assigned students for an intention-to-treat analysis, even if some do not use the offer. Block assignment by relevant deadline window and baseline preparation where feasible. Predefine procedures for urgent needs and record crossover. If delaying support could impair a real application, choose an observational comparison or a different design rather than imposing a delay solely for cleaner inference.
How to read a future results table
Imagine three deliberately constructed packets used to train reviewers. The first has a well-sourced university list and a complete calendar, but the financial column assumes an award that has not been offered. The student can explain the scholarship deadline and has a realistic plan to submit. A reviewer should give credit for timing and rule interpretation while flagging the unresolved cost. Giving full credit simply because the student has applied for aid would merge two decisions: eligibility to compete and ability to pay after an uncertain result. Neither a positive attitude nor an attractive list resolves that uncertainty.
The second packet lists one university and one community college pathway. The student explains why commuting is important, identifies the direct-entry rule and documents what would be required to move into the desired program later. The plan is short, inexpensive and personally defensible. If a reviewer assigns a low list score because the student did not include a highly selective institution, the reviewer has imported an objective that the student did not choose. Calibration should include examples like this one so that scores reflect capacity to make an informed choice rather than a counselor’s preferred list shape.
The third packet contains a polished schedule that appears to anticipate every deadline. When asked to interpret an unfamiliar program condition, however, the student cannot say whether the condition applies and does not know where to check. That pattern is consistent with receiving a good calendar while learning less about how to maintain it. It would be inappropriate to accuse the student or counselor of improperly producing the document. It would be equally inappropriate to treat the document as proof of independent competence. A new case and a short teach-back allow researchers to distinguish the delivered artifact from the skill the student can carry to the next application.
These fictional cases are useful precisely because they are not data about anyone’s performance. The public cohorts above already show comparable distinctions between types of observed outcome. In Washington, younger students consistently became more confident about selecting a college, but the evaluation did not score their selected colleges. In Texas, a paired test measured knowledge about scholarship applications, while the separate paragraph-scoring endpoint contained reporting inconsistencies. The California virtual trial found changes in applications that did not establish gains in admission or enrollment. A local director who wants to know whether the service helped should name which kind of evidence would answer the question before inspecting the results.
A publishable results table would begin with counts, dates and the unit of observation. It would identify the number of students invited, the number consenting, and the number with paired work, followed by outcomes in the prespecified order. Where the local result is missing, the table would display an explicit unavailable label, never a row of zeros. Where external results are included, they should be grouped by intervention and endpoint with their own sources. An external benchmark can put a result in context without serving as a control group. In particular, an old residential summer program is an informative comparison of measurement practices, not a substitute for contemporary students who did not receive a particular commercial service.
For a family, the most useful question after a month of advising may be very concrete. Can the student identify a financially workable alternative if the preferred plan fails? Can they explain which deadline governs a scholarship or portfolio? Can they independently verify a changed requirement? Can they distinguish an item they have submitted from an item the college has actually received? Observing those capabilities is possible without asking the student to attribute every achievement to a consultant. It is also possible before the first admission letter arrives.
For a program evaluator, the corresponding responsibility is to publish findings that could be unwelcome. A rise in confidence alongside a flat knowledge score should appear in the abstract if it changes the interpretation. An increase in application breadth with no clear improvement in affordability should remain visible. A drop in scores after a change to a more difficult assessment should trigger examination of form equivalence. If students who are hardest to reach are missing, that fact belongs in the account of whose experience the study represents. Success cannot be defined retrospectively as whatever happened to improve.
The available evidence answers the title conditionally. Structured programs have improved particular application actions, tested knowledge and perceived preparation in defined settings. The reviewed work does not support a single effect size for all forms of advising or a claim that more readiness guarantees an admission offer. The strongest next contribution would be a public set of genuinely comparable measurements, obtained with appropriate consent, that shows what students could demonstrate before and after a clearly defined process. Until such measurements exist for a named provider, its own effect remains an open empirical question.
Disclosure and data availability
Prepared for The Ivy Institute using published external evidence. No private student records or client testimonials were used as research observations, and no individual students or families were contacted. The accompanying records identify original tables, denominators, scales and calculations. The blank cohort template and scoring manual are prospective research instruments. They describe how a future study could be run with appropriate consent and independent assessment.
References and primary records
- R1 · Oreopoulos and Ford (2019). Keeping College Options Open. Source record.
- R2 · Phillips and Reber (2022). Does Virtual Advising Increase College Enrollment?. Source record.
- R3 · Phillips and Reber, NBER Working Paper 26509. Numerical extraction from Tables 5–8 of the public working paper. Source record.
- R4 · Bettinger, Long, Oreopoulos and Sanbonmatsu (2012). The Role of Application Assistance and Information in College Decisions. Source record.
- R5 · Carrell and Sacerdote (2017). Why Do College-Going Interventions Work?. Source record.
- R6 · Bird and colleagues (2021). Nudging at Scale: Experimental Evidence from FAFSA Completion Campaigns. Source record.
- R7 · College Guidance for All (2019). School-level randomized evaluation of near-peer advising. Source record.
- R8 · Barr and Castleman (2025). Increasing Degree Attainment among Low-Income Students. Source record.
- R9 · Harvard DASH research repository. The Effects of College Counseling on High-Achieving, Low-Income Students. Source record.
- R10 · U.S. HHS Office for Human Research Protections. Research with Children FAQs. Source record.
- R11 · Clayton and Worsham (2024). Preparing Students for Postsecondary Success. Source record.
- R12 · Washington State Institute for Public Policy (2008). Make It Happen! Preliminary evaluation, Exhibits A.1–A.3. Source record.
- R13 · Murr (2010). A Scholarship Workshop Program to Improve Underrepresented Student Access to Higher Education. Source record.
- R14 · Early and DeCosta-Smith (2011). Making a case for college: genre-based essay instruction. Source record.
- R15 · Phillips and Reber, American Economic Association data archive (2022). Replication archive and documentation, version 1. Source record.
- R16 · National Center for Education Statistics. High School Longitudinal Study of 2009, public data documentation. Source record.
- R17 · Cooper, Gilles, Bautista, Alvarez and Saroha (2018). My Journey to College, UCLA research poster. Source record.
- R18 · Amaro-Jiménez, Pant, Hungerford-Kresser and den Hartog (2020). Identifying the Impact of College Access Efforts on Parents’ College Preparedness Knowledge. Source record.
- R19 · Martinez and colleagues. Helping Diminish the College Knowledge and Access Divide. Source record.
- R20 · USC Pullias Center (2016). From Pre to Post: Putting the Data Pieces Together. Source record.
