In the second week of September we compared the foundational library, course by course, against the research-methods catalogue of a commercial training provider. Six of their courses had no counterpart here. All six are now live as free 101 decks: one hundred slides each, about 122,000 words between them, no login and no fee. This post says what each one is for and where it takes a position.
The six sit on the research side of the house rather than the programme side. They are for the PhD scholar in a state university, the research officer in an NGO who has been handed a systematic review, and the M&E lead asked for something beyond the monitoring report. The 52 decks that already existed covered the sector and its substance. These cover the craft: how to find what is known, how to write it up, how to model a series, a scale, a survey or a set of interviews, and how to do so without a programming language if that is where you are.
Each course was built the same way. Every worked table is labelled illustrative, because a made-up dataset presented as a real one is a lie with decimals. Every empirical claim on a slide was traced to a primary source before it went in, and several were corrected on the way. And each course says, in plain words, what it cannot do.
Systematic reviews and evidence synthesis
Systematic Reviews & Evidence Synthesis 101 covers the whole chain: choosing between a systematic, scoping, rapid or realist review; writing and registering a protocol; searching three databases and the grey literature and saving the strategy as run; screening in pairs; extracting data and converting reported statistics into effect sizes; risk of bias with RoB 2, ROBINS-I and the quasi-experimental tools; a meta-analysis worked by hand under fixed and random effects; synthesis without pooling; GRADE and the summary-of-findings table; PRISMA 2020; and a bibliometric map of a field in VOSviewer and Bibliometrix. Every tool it names is free.
The position it takes is that the boring safeguards are the evidence. Buscemi and colleagues (Journal of Clinical Epidemiology, 2006) compared single and double data extraction in the same review and found that a single extractor made 21.7 per cent more errors, relative to the double-extraction rate, and saved 36 per cent of the time. The course puts those two numbers side by side and asks which one your funder will remember. On why the grey literature search is not optional, it uses Franco, Malhotra and Simonovits (Science, 2014), who followed 221 social-science studies from a single funded programme: of the studies that produced null results, 10 of 48 were published, against 56 of 91 with strong results, and roughly two-thirds of the nulls were never written up at all. A review that searches only the published record is searching a sample selected on its result.
Academic writing and publishing
Academic Writing & Publishing 101 starts with the claim before the paper and works outward: the hourglass shape of an argument, IMRaD and its economics and development-studies variants, a literature review that argues rather than lists, sentences with characters and actions, numbers rounded in prose and exact in tables, citation styles, the UGC's plagiarism tiers, the PhD regulations, the viva, choosing a journal, open access, preprints, the point-by-point response to reviewers, authorship under the ICMJE criteria, and what to disclose about AI. Every rule comes with a before-and-after.
Three regulatory facts shape the Indian half of it, and all three changed recently enough that older guides get them wrong. The UGC PhD Regulations of 2022 removed the requirement to publish a paper before submitting the thesis, so the pressure to place something, anything, in a journal before the viva has no rule behind it any more. The UGC-CARE journal list was discontinued on 11 February 2025 and replaced with a set of suggestive parameters, eight heads and 36 items, which means the question "is it on the list" has no answer and a scholar has to judge a journal for themselves; the course teaches how. And One Nation One Subscription went live on 1 January 2025, giving around 6,400 government institutions access to more than 13,000 journals from 30 publishers, so the paywall excuse for a thin literature review is weaker than it was. On the other side of the desk, Card and DellaVigna (Journal of Economic Literature, 2013) documented acceptance rates at the top five economics journals falling from about 15 per cent to about 6 per cent over three decades, which is the context for the course's advice to aim one tier down and revise fast.
Time series analysis
Time Series Analysis 101 is built for applied work with Indian data: plotting before modelling, trend and seasonal decomposition with STL and X-13, unit-root tests read as a table rather than a verdict, ARIMA identified step by step, exponential smoothing and ETS, spurious regression and the cointegration that rescues levels, the ARDL bounds test done properly, VAR impulse responses and what Granger causality does not mean, GARCH and structural breaks including 2016 and 2020, and the three designs for evaluating a programme with no control group: interrupted time series, synthetic control and CausalImpact. It closes with where every Indian macro and administrative series lives and what each costs to use.
Its one insistence is that 120 monthly observations are not 120 observations. Autocorrelation eats degrees of freedom, and the course shows a regression of one trending series on another that reports a t-statistic above 10 for a relationship that does not exist. Forecasts are judged out of sample against the seasonal naive benchmark, because a model that cannot beat "the same month last year" has not earned its complexity.
Structural equation modelling
Structural Equation Modelling 101 is about measuring what no single question can: what a latent variable is and is not, reflective against formative constructs, confirmatory factor analysis read from real output, omega, AVE and HTMT as a validity table rather than a pass mark, identification by counting, WLSMV for the binary and three-category items that South Asian surveys are made of, FIML for missing data, the chi-square and the fit indices with judgement, mediation with bootstrapped indirect effects, measurement invariance across Hindi and Bangla versions of a scale, PLS-SEM with the case for and against it, common method bias, equivalent models, and the verbs each design allows.
The slide most readers will argue with is the one on modification indices. Freeing a parameter because the software says fit would improve produces a model that fits this sample and means nothing about the next one, and the course shows the same data yielding two incompatible "well-fitting" models by that route. The invariance chapter matters for anyone translating an instrument: a scale that measures a different construct in Bangla than in Hindi cannot be used to compare the two groups, and the test for that is cheap to run and rarely reported.
Statistics without code
Statistics Without Code 101 does real statistics in jamovi and JASP, the free point-and-click tools built on R. Every analysis slide has the same shape: the question, the menu path, the boxes to tick and why, an illustrative output table, what to do when the assumptions fail, and the sentence for the report. Two illustrative datasets run through the whole course, a self-help-group survey of 640 women in 32 villages and a learning assessment of 1,200 children in 40 schools, so the reader sees the same evaluation analysed naively and then properly. In the worked example the cluster-adjusted confidence interval is twice the width of the naive one.
One claim was corrected against jamovi's own documentation before the course shipped. jamovi does accept a weights variable, but only integer frequency weights, with no strata and no clusters, and JASP has no survey-design support at all. So a weighted national survey such as NFHS or PLFS still needs R's survey package or Stata's svy commands, and the course gives the four lines of R that do it rather than pretending the menus can.
Qualitative analysis software
Qualitative Analysis Software 101 covers NVivo, MAXQDA, ATLAS.ti and the free tools Taguette and QualCoder: what the software does and does not do, consent and where the transcripts may go, transcription in Hindi, Bangla and Tamil with a native-speaker check, three translation workflows and what each loses, codebooks with definitions, first and second cycle coding, memos as the place the analysis happens, retrievals read before anything is counted, matrices as pointers rather than proportions, coding in teams with agreement measured per code, joint displays that make interviews and a survey one study, COREQ and SRQR, and the AI features: what they do, where the data go, how to validate them like a second coder, and the disclosure sentences.
The free tools come first and in more detail, because they are the ones most readers can install today. One illustrative study, 36 interviews with SHG members in Bihar and West Bengal, runs through the whole course: twelve interviews coded by one analyst in Taguette, then all 36 across two languages and two coders in QualCoder with agreement checked, and only then the commercial three compared. It connects to two things already on the platform, VaniScribe for Indic transcription and Data Protection & the DPDP Act 101 for the consent questions that arise the moment a recording leaves the phone it was made on.
What they have in common
All six are native HTML decks in the 101 series, with light and dark themes, keyboard navigation, and a table of contents you can jump around. The library now stands at 58 foundational courses. Two of the new ones are filed under Data & Technology, the other four under MEL & Research, and all six are in site search and the catalog.
They also share a limit worth stating. A hundred slides can teach the shape of a method and the mistakes that matter; it cannot replace doing it once with your own data and someone who has done it before looking over your shoulder. Each course ends with a practice section for exactly that reason, and each names the textbooks and documentation it drew on so the reader knows where the next hundred pages are.
What is next
Almost everything on ImpactMojo is in English, and that has been the open item longest on the list. In September 2026 Bodhan AI and AI4Bharat released open-weight models for speech to text, text to speech, OCR and translation across the 22 scheduled languages. They are now on the roadmap, with transcription that stays on the user's machine as the first planned use, which is also the use the qualitative software course was written around.
If you open any of these six and it fails you somewhere, a figure that does not reconcile, a menu path that has moved in a newer version, a slide that assumes something it should not, that is worth telling us. The material improves mostly when somebody reports where it broke.