home>Nature Methods Points of Significance

statistics + data

The Nature Methods Points of View column column offers practical advice in design and data presentation for the busy scientist.

50+ columns of good explanations ·

With the publication of Uncertainty and the Management of Epidemics, we celebrate our 50th column! Since 2013, our Nature Methods Points of Significance has been offering crisp explanations and practical suggestions about best practices in statistical analysis and reporting. To all our readers and coauthors: thank you and see you in the next column!

Nature Methods: Points of Significance

Martin Krzywinski @MKrzywinski mkweb.bcgsc.ca

▲ Points of Significance column in Nature Methods. (Launch of Points of Significance)

≡ authors figures bibliography metrics

Generated on 10-Sep-2026 (0 days ago).

Metrics are provided by Altmetric.

Access values larger than 10,000 are rounded off to nearest 1,000 by Altmetric.

article	accesses	daily	%nm	%all	citations	altmetric
Error bars	236,000	50	98	99	282	247
Significance, P values and t-tests	234,000	50	90	98	168	85
Principal component analysis	217,000	65	86	96	1,428	76
Visualizing samples with box plots	217,000	47	88	98	539	82
Association, correlation and causation	212,000	53	98	99	368	190
Replication	144,000	33	63	95	212	30
Statistics versus machine learning	141,000	46	97	99	1,475	346
P values and the search for significance	137,000	39	90	98	109	128
Importance of being uncertain	127,000	27	79	97	93	52
Power and sample size	116,000	25	80	97	213	63
The SEIRS model for infectious disease dynamics	87,000	38	98	99	230	363
Nonparametric tests	68,000	15	58	89	101	14
Comparing samples—part I	67,000	15	64	92	71	21
Analysis of variance and blocking	66,000	15	60	91	75	18
Two-factor designs	63,000	15	34	51	36	3
Bayes' theorem	62,000	15	69	94	108	33
Simple linear regression	59,000	15	77	94	162	35
Sources of variation	59,000	14	48	88	33	12
Comparing samples—part II	58,000	13	48	86	82	11
Split plot design	57,000	14	40	77	72	7
Classification and regression trees	52,000	16	45	80	389	9
Multiple linear regression	52,000	13	80	96	166	43
Sampling distributions and the bootstrap	52,000	13	65	93	169	23
Designing comparative experiments	52,000	12	52	81	21	8
Bayesian statistics	51,000	12	51	88	37	14
Interpreting P values	50,000	14	70	95	80	48
Bayesian networks	45,000	11	39	73	73	6
Model selection and overfitting	44,000	12	75	95	676	42
Classification evaluation	44,000	12	67	94	414	40
Clustering	38,000	11	52	90	156	21
Optimal experimental design	31,000	10	51	90	123	21
Analyzing outliers: influential or nuisance?	29,000	8	29	65	115	4
The curse(s) of dimensionality	27,000	9	60	92	498	26
Machine learning: supervised methods	25,000	8	60	93	342	29
Logistic regression	25,000	7	66	92	129	26
Modeling infectious epidemics	24,000	10	86	97	105	114
Regression diagnostics	24,000	6	46	72	81	6
Ensemble methods: bagging and random forests	21,000	6	40	85	316	13
Machine learning: a primer	20,000	6	86	97	261	71
Tabular data	14,000	4	23	49	8	3
Regularization	14,000	4	57	88	46	17
Two-level factorial experiments	13,000	5	12	24	20	1
Markov models—Markov chains	10,000	4	36	89	36	19
Predicting with confidence and tolerance	9,065	3	23	46	10	3
The standardization fallacy	8,450	4	41	89	55	17
Markov models — hidden Markov models	7,329	3	28	77	37	8
Convolutional neural networks	7,077	6	18	73	103	6
The class imbalance problem	5,574	3	31	85	138	12
Quantile regression	5,427	2	35	84	115	12
Survival analysis—time-to-event data and censoring	5,288	4	20	70	32	5
Markov models — training and evaluation of hidden Markov models	4,867	2	44	83	11	10
Analyzing outliers: robust methods to the rescue	4,255	2	25	50	35	3
Neural networks primer	4,059	3	21	69	11	4
Propensity score matching	4,048	6	5	31	36	1
Propensity score weighting	2,876	5	18	71	10	5
Graphical assessment of tests and classifiers	2,639	1	25	80	11	9
Regression modeling of time-to-event data with censoring	2,543	2	16	72	6	5
Comparing classifier performance with baselines	2,424	3	24	76	13	7
Errors in predictor variables	1,955	2	1	14	7	1
Testing for rare conditions	1,862	1	15	60	4	4
Double robustness	1,484	11	1	1	0	1
Symmetric alternatives to the ordinary least squares regression	1,347	3	37	78	1	7
Consequences of missing data	440	63			0
Uncertainty and the management of epidemics	0	0			0
Nested designs	0	0			0
	3,267,009	946	48	76	10,753

%NM percentile rank (avg 48) for the article of tracked articles of a similar age in Nature Methods.

%ALL percentile rank (avg 76) for the article of tracked articles of a similar age in all journals.

Total accesses 3,267,009.

Total citations 10,753.

VIEW ALL

news + thoughts

Consequences of missing data

Fri 04-09-2026

Either I've been missing something or nothing has been going on. —Karen Elizabeth Gordon

Missing data are everywhere. Subjects may decline to participate in a survey or fail to answer sensitive questions. A cell culture might fail due to contamination. Instrument failures or mishandling of a sample may lead to missing observations. But why are missing data a problem?

▲ Nature Methods Points of Significance column: Consequences of missing data. (read)

This month, we begin a series of articles about practical and statistical aspects of missing data. We'll see that missing data can increase variability and introduce bias and we will ask whether anything can be done to mitigate these consequences. It turns out that, in the case where we know nothing about the missing subjects, no mitigation is possible. We must accept higher variability and, if we have a suspicion that the missing subjects arenât completely random, possible bias as well.

Tanujit Dey, T., Lipsitz, S.R., Fitzmaurice, G., Krzywinski, M. & Altman, N. (2026) Points of significance: Consequences of missing data. Nat. Methods 23:1661–1663.

Double Robustness

Mon 04-05-2026

It is not certain that everything is uncertain. —Blaise Pascal

We have already explored how we can mitigate bias caused by confounding variables in observational studies using propensity score (PS) matching (PSM) and propensity score weighting (PSW). However, any statistical model is only as good as its assumptions and, if it is specified incorrectly, it can itself produce biased estimates of the treatment effect.

▲ Nature Methods Points of Significance column: Double Robustness. (read)

This month, we explore double robustness, a powerful statistical concept that provides a valuable “safety net” against the risk of an incorrect model. It offers two opportunities, instead of just one, to obtain a valid estimate of the treatment effect â making it possible to draw credible causal inferences from observational data without having to depend on a single set of modeling assumptions.

Kurz, C.F., Krzywinski, M. & Altman, N. (2026) Points of significance: Double Robustness. Nat. Methods 23:868–869.

Nature Biotechnology cover

Thu 23-04-2026

My cover design on the 7 April 2026 Nature Biotechnology issue shows the dendrogram that represents a cluster of uniquely expressed (or downregulated) genes in human naive stem cells induced from such cells. Within each dendrogram block, the genomic barcode sequence (sampled from Supplementary Table 1) is depicted with a Code 39 barcode. The highlighted barcode is one of those used for cell isolation.

Ishiguro S. et al. A multi-kingdom genetic barcoding system for precise clone isolation (2026) Nature Biotechnology 44:616–629.

▲ My Nature Biotechnology phylogenetic tree cover (volume 44, issue 4, 7 April 2026). (more)

Browse my gallery of cover designs.

▲ A catalogue of my journal and magazine cover designs. (more)

Happy 2026 Ï Day—
Art for the 5%

Fri 13-03-2026

Celebrate Ï Day (March 14th) and enjoy the art — but only if you're part of the 5%.

Go ahead, see what you can't see.

▲ 2026 Ï DAY | Art for the 5%. Shown in the style of Ishihara color test plates, the art is visible only to those with colour blindness. (details)

Nature Methods: Points of Significance

Consequences of missing data

Double Robustness

Nature Biotechnology cover

Happy 2026 Ï Day—Art for the 5%

Happy 2026 Ï Day—
Art for the 5%