Two Months to Submission and You Can't Code: R, SPSS or jamovi?

Your data are collected, submission is eight weeks away and your results chapter is empty. You ask which package to learn and get dozens of replies arguing with each other: R is the standard and everything else is for amateurs, SPSS belongs to another decade, jamovi is a toy. None of them answers what you were actually asking, which is not which package is best but which one gets you finished on time with something you can defend — in a viva, in front of a committee, or in an examiner's written report. That question does have an answer, and it turns on two things: what analysis your design demands, and how many weeks are genuinely left once you subtract the writing.

The real question is not which is best, it is which one finishes on time

Software comparisons are written for someone choosing a package at the start of a PhD with four years ahead of them. You are not that person. The criteria change completely: not which is most powerful or most used in your field, but which one puts a correct result on screen soonest, and which one leaves you able to explain what came out. If you want the cool-headed comparison of all three, with their underlying strengths and no clock running, that is in SPSS, R or JASP: which to choose. This is the version with a deadline in the room.

Under deadline the decision comes from three numbers, not from anyone's opinion. First, what analysis your design demands — not what you would enjoy running, what the design obliges you to run. Second, how many weeks remain after you subtract writing, formatting and administration. Third, who you can ask when it jams, which is the factor no comparison mentions and the one that decides the most weeks: a supervisor who opens your output and spots the error in two minutes is worth more than any technical advantage of a package nobody in your department uses.

That gives you the only ordering rule that matters: choose the tool after the analysis plan, never before. Writing the plan — which hypotheses, which variables, which test for each, what you do when an assumption fails — costs an afternoon and prevents the classic disaster of these eight weeks, which is spending three of them learning a package for an analysis that turns out not to be yours. If you do not have that document written down, start with the thesis data analysis plan and come back with it in front of you.

The slow part is the model, not the software

There are two learning curves and they get confused constantly. One is the package: where you click to get a repeated-measures ANOVA. The other is the model: what a violation of sphericity means, why the Greenhouse-Geisser correction turns your degrees of freedom into decimals, and which of the contrasts the table spits out actually tests your hypothesis rather than a different one. The first curve takes an afternoon with a decent tutorial. The second does not, and it does not care which software you picked.

When the analysis works, it looks like this: in a 2 × 3 design with N = 84, the group × time interaction gives F(2, 164) = 4.31, p = .015, η²_p = .050. Producing that line takes forty minutes in any of the three packages. What you will defend is not the line: it is why the effect that tests your hypothesis is the interaction rather than the main effect of time, and what η²_p = .050 (which sits between small and medium: the convention in circulation since Cohen, 1988 puts .01 at small and .06 at medium, and those are conventions, not laws) means in your particular literature. No menu solves any of that.

This is why the wrong tool gets paid for twice. Once in the hours spent fighting it. And once, far more expensively, in the hours those hours steal from thinking about the model, which is the only thing your examiners will interrogate. With two months left, every hour spent installing dependencies is an hour not spent working out why your effect vanishes when the covariate goes in.

The practical consequence is simple and thoroughly unglamorous: if your analysis plan fits inside the menus, use the menus. Elegant syntax earns no marks at the viva, and a correct analysis produced by clicking is exactly as valid as the same analysis written in code.

jamovi: the default answer when your plan is standard

jamovi is free, installs in ten minutes, opens SPSS .sav files with their labels intact along with .csv and .xlsx, and is R underneath: when you click an ANOVA, the engine running is the one that would run in a console. That detail answers the toy accusation that appears in every thread. The same model returns the same number everywhere: a regression with three predictors and N = 120 gives R² = .34, F(3, 116) = 19.92, p < .001 wherever you run it. What changes is not the result, it is who writes the syntax.

What it does with no code at all. Descriptives and plots; all three t tests with their non-parametric alternatives (Mann-Whitney, Wilcoxon); factorial, repeated-measures and mixed ANOVA plus ANCOVA, with corrected post hoc tests and effect sizes; full correlation matrices; linear and logistic regression (binomial, multinomial and ordinal); contingency tables and chi-square; reliability with Cronbach's α, McDonald's ω and item-rest correlations; exploratory and confirmatory factor analysis with fit indices.

What it does with a module, two clicks and still no code. GAMLj for mixed and generalized linear models; medmod and jAMM for mediation and moderation with bootstrap intervals; MAJOR for meta-analysis; Rj for pasting in R code the day one odd thing needs it. The module library sits inside the program, so nothing has to be hunted down online.

Where the ceiling is. Multiple imputation is not solved from a menu, large structural equation models need the syntax module and start to feel like writing code anyway, and unusual designs — three-level nesting, latent growth curves — fall outside the catalog. The full account of what is in and what is out, with the workflow for a complete thesis, is in jamovi for your thesis.

SPSS: the one you already have, plus two crutches worth weeks

If your university holds a site license, you saw it in your methods course and your supervisor reads its output, SPSS starts with an advantage that has nothing to do with statistics: there is someone down the corridor to ask. Its curve is middling — menus, yes, but with its own vocabulary and output you have to learn to read — and having help nearby more than covers that friction when eight weeks remain. Check the obvious thing today: that your student license runs past your submission date, because they expire, and they expire badly.

Crutch one: PROCESS. Hayes's macro installs as a dialogue box and runs mediation, moderation and conditional process models with bootstrap confidence intervals without a line of code. It is the reason a large part of social psychology is still on SPSS. Its limit: it works with observed variables, not latent ones. If your model needs latent variables you are in AMOS or R territory instead.

Crutch two: mixed models from a menu. If your design is longitudinal and you lost people along the way — you recruited 120 and only 98 completed all three waves — repeated-measures ANOVA deletes 22 whole participants for having one gap, while a mixed model keeps them with the measurements they did give. SPSS runs those from Mixed Models with no code, and it genuinely changes your N. Why this is not a reviewer's whim is set out in mixed models versus repeated-measures ANOVA.

What SPSS will not give you. Confirmatory factor analysis and structural equation modeling, which live in AMOS — a separate product with a separate license your institution may not hold. Multiple imputation and general bootstrapping sit in add-on modules your license may not include, so verify that before you plan around them. And the output arrives in tables that have to be rewritten wholesale for APA, an afternoon worth budgeting into the calendar.

R in two months: when it works and when it is a trap

The R curve is not the curve of an analysis, it is the curve of an environment. You will run your first t test in R on the first afternoon. What you will not acquire in two months from a standing start is the ability to unstick yourself when the cryptic error turns up at eleven on a Thursday night and the message talks about an object of length zero. That moment decides whether R saves you or sinks you, and under deadline you would be learning two things at once — the model and the language — with no way of telling which one is failing when the output looks wrong.

There is one case where R stops being an option and becomes the only door: when the analysis your design demands exists in no menu anywhere. Structural equation models with ordinal indicators and a WLSMV estimator (lavaan), three-level or cross-classified multilevel models (lme4), item response theory (mirt), meta-analysis with moderators (metafor), latent growth curves. If your thesis needs one of those, the real decision is not between packages: it is between learning R in time and commissioning that analysis. Make it now, not in week seven.

There is a middle road almost nobody uses and it suits you: jamovi will show you the R syntax for whatever you just clicked. Turn on syntax mode, run the analysis with the mouse, copy the code that appears and file it with your supplementary materials. You come out with a reproducible analysis and the code in hand without having programmed — which is exactly what a journal with an open-science policy will ask for.

In case this sounds like R is beneath you: learning R remains the best investment of time available if your career continues after this. It simply does not happen in the eight weeks before submission, any more than you learn to drive on moving day. It happens afterwards, on data with no deadline attached.

When the right answer is to delegate rather than learn

Delegating is not surrender, it is arithmetic. Count the weeks left, subtract writing the results and discussion, subtract formatting, subtract the time your supervisor needs to read and return a draft, and look at what remains for analysis. If the answer is three weeks and your plan contains a model you have never fitted in a package you have never opened, this is not a problem of effort or attitude: it does not fit. Far better to know that today than to discover it on day 12 of the 15 you had.

Four signals beat intuition here. One: the analysis your design demands is in no menu of any package you can use. Two: you have made two attempts and the error was different each time, which means you are not debugging, you are guessing. Three: the block is not running the analysis but choosing which analysis to run, and no tutorial resolves that because it depends on your specific design. Four: it runs, but you could not say out loud why that model and not another.

The ethical line is exactly one and worth writing down: you can commission someone to run and document an analysis, you cannot commission someone to understand on your behalf. You are the one being examined, and if you cannot explain every analytic decision as though you had made it, you have not delegated, you have outsourced. Where the line sits and what most universities allow is worked through in is it ethical to hire a statistician for your thesis.

Delegating well has a concrete shape: you ask for the analysis and the syntax that produced it, you ask for APA-formatted tables, and you ask for a session where every decision is explained until you can restate it in your own words. If what comes back is a PDF of results with no code and no explanation, that is not consulting: it is a timer set to go off in the viva, the moment someone asks why you chose that estimator.

Eight weeks, counted backwards from the deadline

Count the calendar backwards from the deadline, never forwards from today, because forwards everything always looks like it fits. Subtract the non-negotiables first: the week of formatting, submission and paperwork, and the two that go on writing results and discussion and getting them back marked up. Eight weeks leaves five real ones for analysis, and one of those goes on work that does not feel like analysis at all: cleaning the file, recoding, reversing the items that run backwards, deciding what happens to the missing values.

A split that works: week 1, analysis plan and data cleaning. Week 2, descriptives, scale reliability and assumptions. Weeks 3 and 4, the main analysis, the one that answers your central hypothesis. Week 5, secondary analyses and robustness checks, with tables being built in parallel from week 2 rather than at the end. If cleaning swallows the whole of week 2 in your case, fine. What cannot happen is that it swallows week 4.

Put a checkpoint at the end of week 3 and honor it: if the main analysis has not run once, even badly, throwing a weekend at it will not fix the problem. That is the moment to change tool, simplify the model or ask for help, and it is the moment when changing something still changes the outcome. By week 7 nothing changes any more; by week 7 the only variable left is how much of the thesis you cut.

Then keep a whole week in reserve, unassigned. It will go — it always goes: a file that will not open, a model that will not converge, a variable miscoded since data collection, a supervisor who reads the final draft and asks for one more analysis. That week is not luxury slack. It is the part of the plan that makes the rest of the plan credible.

Frequently asked questions

Can I do my entire thesis in jamovi?

If your analysis plan is the usual one, yes: descriptives, t tests, ANOVA in all its forms, ANCOVA, correlations, linear and logistic regression, chi-square, reliability with α and ω, exploratory and confirmatory factor analysis, and mediation or moderation with bootstrap intervals once you install a module. It falls short on multiple imputation, on large structural equation models and on three-level nested designs. Read your analysis plan line by line: if every line is in that first list, jamovi will carry you to submission.

How long does it actually take to learn R?

It depends what you mean by learn. Running basic analyses by following a tutorial: two or three weeks. Being self-sufficient when something breaks — reading an error, realizing your variable came in as text rather than a factor, fixing it unaided — takes months, not weeks, and that is the level you need if R is going to be your tool with a submission date in the diary. The honest test is the Thursday-night one: if a cryptic error stops you dead and there is nobody to ask, you are not at that level yet.

My university has no SPSS license, what now?

jamovi and JASP are free, install without administrator rights on most machines, and open your .sav file directly, so nothing you have already done is lost. Between them: jamovi suits the workflow of a whole thesis better (it edits data, computes new variables and stores everything in one file with the analyses inside), and JASP suits you better if you want Bayesian alongside frequentist tests. Do not buy a personal SPSS license for two months of work.

Will my examiners ask which software I used?

Almost never, and when they do it is curiosity or they want the version for your analysis section. Nobody cares whether t(88) = 2.28, p = .025, d = 0.48 came out of a menu or a console. What you will be asked is why you compared those two groups, what you did about the assumptions, why that effect size matters and what happened to the participants who dropped out. Report the package, the version and the modules or libraries used in your analysis section and you have covered it.

Can I move my SPSS data into jamovi without breaking anything?

Yes — jamovi opens .sav files with variable and value labels intact. Check two things on opening, because that is where it breaks: that variable types came across correctly (nominal, ordinal, continuous), since the menus offer analyses based on those, and that user-defined missing values from SPSS (the usual 99 or 999) are still flagged as missing rather than sitting in the file as real scores. One stray 99 treated as data drags the mean of a 0-to-4 scale somewhere unrecognisable.

Is it cheating to have a statistician run my analyses?

No, if you understand what was done and can defend it as your own; yes, if a result arrives that you cannot explain. The practical difference is in the deliverable: an analysis with its syntax, its decisions justified and a session to walk you through it leaves you able to defend; a PDF of tables with no explanation leaves you worse off than before, because now you own numbers you do not control. Check your institution's rules on external help as well — most allow it, and some ask that it be declared in the acknowledgments.

Hold on to the ordering, because that is what breaks in these eight weeks: the analysis plan first, the tool second, and the smallest tool that does what the plan requires. jamovi by default, SPSS if you already have it and there is someone nearby to ask, R only when your obligatory analysis lives nowhere else. And the decision to ask for help belongs in week 3, while something can still change, not in week 7, when the only thing left to change is how much of the thesis survives.