189,480 dreams. 25,420 journals. One question: what returns when we sleep?
A multilingual content analysis, weighted so every journal counts equally. No individual report is published.
What all our analyses confirm
Home and family rank first and second however the data are counted.
Home or building ranks #1 by raw volume, by the equal-journal estimate and after removing the most prolific journals. Family remains #2 in the same analyses.
What appears in the average journal’s dreams
Every journal receives equal weight, whether it contains one dream or several thousand. The bar is the average share of a journal’s reports containing the theme; the line is a 95% confidence interval from 2,000 journal-level bootstrap samples.
Raw rankings can tell a different story
Transport or travel ranks third by raw report volume but fifth when every journal carries equal weight. The gap reflects highly concentrated contributions, so both views belong in an honest report.
After removing the largest journal, then the largest 1% of journals, home remains #1 and family remains #2.
Themes that travel together
Lift compares the observed pair frequency with the frequency expected if the themes were independent. A lift of 2 means the pair appears twice as often as expected.
2.2×17,690 dreams · times the expected frequency
1.6×10,230 dreams · times the expected frequency
1.3×9,950 dreams · times the expected frequency
1.6×9,800 dreams · times the expected frequency
1.3×6,520 dreams · times the expected frequency
2.4×6,080 dreams · times the expected frequency
What the corpus looks like
Report length varies widely and contributions are highly concentrated. Both facts shape any conclusion based on raw frequencies.
Report length
Estimated languages
Language is estimated locally from lexical markers. Short or ambiguous texts remain in an uncertain category.
Method and limitations
218,240 exported rows; 28,720 deleted entries and 40 empty reports were excluded.
We calculate proportions separately in every journal and average them. Every journal therefore has equal total weight.
Sixteen multilingual families of words and phrases are searched. One report can belong to several themes.
The 95% intervals come from 2,000 journal-level bootstrap samples with a deterministic seed.
No text, name, identifier, UUID, URL or individual example is exported or published. Very rare categories are suppressed.
This is a Dreamly usage corpus, not a representative sample. Lexical detection can miss a theme or return a false positive. Results describe mentions, not psychological meaning.
Quality controls
189,480 reports reconciled0 duplicate dream IDs0 invalid dates2,000 bootstrap replicates
What the data support
In this corpus, familiar places and close relationships are more widespread than motifs made famous online. Teeth occur in 2.1% of reports and flying in 1.3%. This study measures what users wrote; it does not assign universal meanings to symbols.
References and further reading
- Domhoff, G. W. & Schneider, A. (2008). Studying dream content using DreamBank.net.
- Foglia et al. (2020). Our dreams, our selves: automatic analysis of dream reports.
- Dreamly editorial policy · Terms of use.
