Most universities bar under-eighteens from BSL-2, radioactive and animal work, require constant direct supervision and want safety paperwork weeks ahead. That is why your emails go unanswered, and it is not about you. Economics, epidemiology, political science, linguistics, psychology and computing all support genuine original work with public data and no equipment at all.
Students email thirty labs, hear nothing, and conclude they are not good enough. The actual explanation is written down in university policy documents, and once you have read one the whole problem looks different.
Why labs say no
Because in most cases they are not allowed to say yes.
UNC's policy on minors in laboratories prohibits under-eighteens from working with BSL-2 biological agents, radioactive materials or animals, requires constant direct supervision rather than the ordinary autonomy a lab runs on, and requires safety documentation filed roughly a month ahead. Most research universities have rules of the same shape.
Read that as an operations problem rather than a verdict. Taking a sixteen-year-old means a named supervisor physically present for every hour, paperwork filed before the student arrives, and a set of techniques simply off the table. For a lab running on grant deadlines, that is a real cost against an uncertain benefit.
The honesty of the few labs that publish their position is worth seeing. The Goldhaber-Gordon lab at Stanford receives "tens of such requests per year," takes a high school intern "only once every 4 to 5 years," and tells students "don't be surprised or offended if you don't get a reply."
Once every four to five years. Against tens of requests annually. If you have sent thirty emails and heard nothing, you have observed the base rate, not a judgement on your work.
Which fields genuinely work without equipment
The constraint is data access, not supervision — and in these fields the data is sitting in public.
| Field | What the raw material is | A question shape that works |
|---|---|---|
| Economics | Census microdata, labour and price series | Does X policy variation across states or years track Y outcome? |
| Public health and epidemiology | Mortality, natality and surveillance files | Did areas adopting X earlier show different curves? |
| Political science | Election returns, survey waves, legislative records | Did vote shift differ by Z after controlling for education? |
| Psychology | Repeated national adolescent surveys | Does self-reported A predict self-reported B, controlling for C? |
| Linguistics | Corpora, subtitle and social text archives | How has usage of a construction shifted across decades? |
| Sociology and urban studies | Local authority data, housing and transport records | Do neighbourhoods with X show different Y after deprivation controls? |
| Computer science | Public repositories, APIs, your own logged data | Can a simple classifier do X, and where does it fail? |
| History | Digitised archives, newspaper databases | How did coverage of X change across a defined period? |
The common feature is that somebody has already done the expensive part — collecting data at a scale no school student could — and published it. Your contribution is the question nobody thought to ask of it, which is a real contribution and the one most within reach.
Where to actually find those datasets is a longer answer, and we have written it up separately in where to find a free dataset you can actually use.
If your subject really is wet-lab
Move sideways rather than giving up.
Nearly every laboratory science has a computational wing, and it is usually where the field is growing fastest:
- Microbiology → epidemiology. Same organisms, population-level data, no containment requirement.
- Bench genetics → bioinformatics. Public sequence databases are enormous, free, and underused by students.
- Pharmacology → published-trial meta-analysis. Trial registries and published results, synthesised properly.
- Ecology → citizen-science records. Long-running phenology and bird-survey datasets, decades deep.
- Chemistry → cheminformatics, or an instrument-free measurement study. Label claims versus published independent measurements, for instance.
You keep the subject. You lose only the equipment you were never going to be permitted to touch.
The one thing to avoid is the project that sounds lab-based but is really a demonstration — growing cress under different lights, titrating something at home. Those are fine as school practicals and they are not original research, because the answer is already in a textbook.
Is computational work taken less seriously?
By one institution, explicitly, and it is worth reading exactly rather than paraphrasing.
Caltech reports that 42% of admitted students submitted research, and specifies that it should come from "a laboratory in a university or research-affiliated organization" accompanied by "a letter of recommendation from site of research." Independent research and internships are directed to a portfolio instead.
That is a routing decision rather than a dismissal — the work still gets read, through a different channel. But it does mean a solo project and a supervised lab placement are handled differently at one highly selective university.
Cutting the other way: Yale asks applicants to state "your specific contributions if the work was done in collaboration with others." A student who designed and ran their own analysis answers that in three words. Someone who spent a summer as the fourth pair of hands on someone else's project has a harder paragraph to write, and the difference shows under questioning.
Most universities publish nothing this precise. Anyone telling you confidently how either route is weighted is guessing, including companies selling research programmes.
What a laptop project actually looks like
Concretely, so you can judge the scale.
You pick a dataset — say a national adolescent survey with a few thousand respondents and several hundred variables. You spend a genuine week in its documentation, which is the part everyone skips and the part that separates a real project from a plausible one. You find two variables whose relationship nobody seems to have reported. You check whether an obvious third factor explains it. It usually does, so you find another pairing. Eventually one holds up.
Then you write four thousand words saying what you found, how confident you are, and what would change your mind.
That is a research project. No reagents, no supervisor signature, no safety training, and nobody can stop you starting this evening.
The failure mode is not technical difficulty. It is that nothing external requires a draft by a particular Thursday — which is the thing a mentor mostly supplies, and which you can also supply yourself if you have ever finished something nobody set you.
Where we fit
The Oxford Centre for Advanced Research runs 1:1 mentorships with PhD candidates and early-career researchers at Oxford, Cambridge, other leading UK universities and Ivy League institutions. The Oxford Scholar Programme is £2,000 — about $2,600 — for ten contact hours over ten to fourteen weeks; the Oxford Publication Fellowship is £3,400, roughly $4,400.
Most of the projects we supervise are exactly the kind described above, because they are the kind a school student can actually complete. We do not provide lab access and would be misleading you if we implied otherwise.
If you can already pick a dataset, read its documentation and hold yourself to a schedule, do this alone and keep the money. Our comparison of mentored and self-directed work sets out the honest test for which you are.
Before you send another email
Stop emailing labs for a week and spend that week in one dataset's documentation instead.
Thirty emails is roughly a 1-in-150 proposition on the published numbers. A week inside a well-documented public dataset reliably produces a question. The second is a better use of the same week.
Frequently asked questions
Why do labs keep saying no?
Usually policy rather than judgement. UNC's laboratory safety manual prohibits under-eighteens from working with BSL-2 agents, radioactive materials or animals, requires constant direct supervision, and requires safety paperwork filed roughly a month in advance. Most research universities have comparable rules, so a professor who wants to take you frequently cannot.
Which fields work without a lab?
Economics, political science, public health and epidemiology, psychology using survey or secondary data, linguistics, sociology, urban studies, history, computer science and statistics. In all of them the data is public, documented, and large enough that genuinely unchecked questions remain.
Is computational work taken less seriously?
Not by journals. Caltech does route things differently in admissions — it specifies research from 'a laboratory in a university or research-affiliated organization' with a letter from the research site, and sends independent work to a portfolio instead. That is a routing decision rather than a judgement on quality, and most universities publish nothing this precise.
What if my field really is wet-lab biology?
Move to the computational corner of it rather than abandoning the subject. Epidemiology instead of microbiology, bioinformatics instead of bench genetics, published-data meta-analysis instead of new assays. You keep the subject and lose only the equipment you were never going to be allowed near.
Do I need to know how to code?
For some of it. A spreadsheet handles a surprising amount, and plenty of good student projects never leave one. If you want to work with large microdata — census extracts, survey waves — then a few weeks of basic Python or R will open far more than it costs you.
Can this get published?
Yes. Student research journals care whether you have original findings, not whether you generated them with a pipette. A clean analysis of public data with a defensible method is a stronger submission than a thin bench experiment, and considerably more likely to be finished.