saturn·

← all readings

wild ghost sightings

source /home/coolhand/html/datavis/data_trove/data/wild/ghost_sightings.csv 10,992 rows 12 columns profiled 2026-05-01 raw JSON static .html .ipynb Report Notebook

Reading

dataset summary · high confidence anthropic:claude-opus-4-7

This dataset catalogs 10,992 reported ghost sightings across the United States, with each record describing a location (city, state, latitude/longitude) and a free-text description of the sighting. Every record is in the United States (country has only 1 unique value), and California, Texas, and Pennsylvania lead the state counts — California alone holds about 9.7% of all sightings. The location column is rich with thematic words like 'school', 'cemetery', 'high', and 'house', hinting at the kinds of places people report hauntings. Worth a closer look: the geographic skew toward a few populous states, and the description field which averages ~70 words per entry and is nearly all unique — a good candidate for text mining. Note that latitude/longitude have ~11.5% nulls and a handful of outliers (including a min latitude of -45 that falls outside the US).

citing: row_count · column_count · country.top_rate · state.top_values · state.entropy_ratio · location.top_words · description.word_mean · description.n_unique · latitude.null_rate · latitude.min · longitude.n_outliers

state · Shows which states report the most ghost sightings — California, Texas, and Pennsylvania dominate.
Show data table
valuecountshare
California10709.7%
Texas6966.3%
Pennsylvania6495.9%
Michigan5294.8%
Ohio4774.3%
New York4594.2%
Illinois3953.6%
Kentucky3703.4%
Indiana3513.2%
Massachusetts3423.1%
Florida3283.0%
Missouri3142.9%
Georgia2892.6%
Wisconsin2742.5%
Alabama2242.0%
Tennessee2212.0%
Washington2182.0%
Oklahoma2111.9%
North Carolina2111.9%
New Jersey1941.8%
location · Distribution of location-name lengths; most are short (median 17 chars) but a long tail goes up to 1,016.
Show data table
charscount
2 – 279633
27 – 531280
53 – 7857
78 – 1037
103 – 1290
129 – 1540
154 – 1790
179 – 2052
205 – 2302
230 – 2560
256 – 2812
281 – 3060
306 – 3320
332 – 3571
357 – 3820
382 – 4081
408 – 4330
433 – 4581
458 – 4840
484 – 5091
509 – 5340
534 – 5600
560 – 5850
585 – 6100
610 – 6360
636 – 6610
661 – 6860
686 – 7120
712 – 7370
737 – 7620
762 – 7880
788 – 8130
813 – 8390
839 – 8640
864 – 8890
889 – 9150
915 – 9401
940 – 9650
965 – 9910
991 – 10161
latitude · Most sightings cluster in the continental US latitudes, but watch for outliers reaching as far south as -45.
Show data table
bincount
-45.02 – -42.231
-42.23 – -39.430
-39.43 – -36.630
-36.63 – -33.831
-33.83 – -31.030
-31.03 – -28.240
-28.24 – -25.440
-25.44 – -22.640
-22.64 – -19.840
-19.84 – -17.040
-17.04 – -14.250
-14.25 – -11.450
-11.45 – -8.6510
-8.651 – -5.8530
-5.853 – -3.0550
-3.055 – -0.25720
-0.2572 – 2.5410
2.541 – 5.3390
5.339 – 8.1370
8.137 – 10.930
10.93 – 13.731
13.73 – 16.530
16.53 – 19.330
19.33 – 22.1391
22.13 – 24.925
24.92 – 27.72176
27.72 – 30.52509
30.52 – 33.32672
33.32 – 36.121609
36.12 – 38.911506
38.91 – 41.712549
41.71 – 44.511854
44.51 – 47.31561
47.31 – 50.11164
50.11 – 52.92
52.9 – 55.72
55.7 – 58.50
58.5 – 61.314
61.3 – 64.094
64.09 – 66.8910
longitude · Longitude spread reveals the east–west footprint of sightings, with a heavy concentration around -87.
Show data table
bincount
-164.7 – -156.486
-156.4 – -148.122
-148.1 – -139.710
-139.7 – -131.43
-131.4 – -12365
-123 – -114.71371
-114.7 – -106.4436
-106.4 – -98.04590
-98.04 – -89.71570
-89.7 – -81.372857
-81.37 – -73.032086
-73.03 – -64.7625
-64.7 – -56.360
-56.36 – -48.030
-48.03 – -39.690
-39.69 – -31.350
-31.35 – -23.020
-23.02 – -14.680
-14.68 – -6.3480
-6.348 – 1.9872
1.987 – 10.320
10.32 – 18.660
18.66 – 26.990
26.99 – 35.330
35.33 – 43.661
43.66 – 520
52 – 60.340
60.34 – 68.670
68.67 – 77.011
77.01 – 85.343
85.34 – 93.680
93.68 – 1020
102 – 110.30
110.3 – 118.70
118.7 – 1270
127 – 135.40
135.4 – 143.72
143.7 – 1520
152 – 160.40
160.4 – 168.71
description · Description lengths are right-skewed: median ~297 characters but reaching up to 5,280 for the most detailed accounts.
Show data table
charscount
2 – 1341623
134 – 2663145
266 – 3982473
398 – 5301528
530 – 662810
662 – 794509
794 – 926303
926 – 1058195
1058 – 1190124
1190 – 132271
1322 – 145363
1453 – 158537
1585 – 171727
1717 – 184920
1849 – 198115
1981 – 211312
2113 – 224510
2245 – 23778
2377 – 25096
2509 – 26412
2641 – 27731
2773 – 29052
2905 – 30371
3037 – 31692
3169 – 33010
3301 – 34330
3433 – 35652
3565 – 36970
3697 – 38291
3829 – 39600
3960 – 40920
4092 – 42240
4224 – 43560
4356 – 44880
4488 – 46200
4620 – 47520
4752 – 48841
4884 – 50160
5016 – 51480
5148 – 52801

Schema

12 columns
Alerts
city text 0.0% 4,385
one_word short_text duplicates
country categorical 0.0% 1
imbalance
description text 0.0% 10,987
near_unique
location text 0.0% 9,903
state categorical 0.0% 51
state_abbrev categorical 0.0% 51
longitude numeric 11.5% 8,774
latitude numeric 11.5% 8,775
city_longitude numeric 0.3% 5,222
city_latitude numeric 0.3% 5,310
location_2 text 11.5% 8,776
allcaps
city_location text 0.3% 5,311
allcaps duplicates

city

text feature one_word short_text duplicates
Short place-name strings (mean 9 characters, 73% one-word) with familiar US city names like Los Angeles, San Antonio, and Houston dominating the top values. Heavy duplication (60%, 6604 rows) is expected for a city field and 4385 uniques suggests broad geographic coverage. Top word frequencies ('city', 'county', 'san', 'st.', 'new', 'fort') confirm conventional US toponymy with no emoji, URLs, or boilerplate noise. Treatment: Treat as a categorical/geographic feature; consider geocoding or grouping rare values before encoding. high · anthropic:claude-opus-4-7
n
10,992
nulls
3 (0.0%)
unique
4,385
len_min
3
len_max
49
len_mean
9.043
len_median
9
len_p95
14
word_mean
1.291
word_median
1
n_empty
0
n_duplicates
6,604
duplicate_rate
0.601
vocab_size
3,988
readability_flesch_mean
20.49
emoji_rate
0
url_rate
9.1e-05
one_word_rate
0.7323
allcaps_rate
0.000455
boilerplate_rate
0

country

categorical metadata imbalance
This column records country, but every one of the 10,992 rows is "United States" — cardinality is 1 and top_rate is 1.0. It carries zero information (entropy 0.0) and cannot discriminate between records. Treatment: Drop; constant column with no signal. high · anthropic:claude-opus-4-7
n
10,992
nulls
0 (0.0%)
unique
1
top_value
United States
top_rate
1
cardinality
1
entropy
0
entropy_ratio
0

description

text free_text near_unique
Free-text descriptions, averaging 70 words (median 55) and reaching up to 5,280 characters, with a Flesch readability of ~69.7 indicating fairly accessible prose. Nearly every row is unique (10,987 of 10,992) with only 5 duplicates and a vocabulary of 33,001 tokens, so this reads as bespoke long-form copy rather than templated text. Boilerplate (0.6%), URLs (0.5%), all-caps (0.06%) and emoji (0%) are all negligible, and the top words are common English stopwords — consistent with natural English narrative. Treatment: Tokenize and embed before modelling; do not use as a categorical key. high · anthropic:claude-opus-4-7
n
10,992
nulls
0 (0.0%)
unique
10,987
len_min
2
len_max
5,280
len_mean
380
len_median
297
len_p95
954
word_mean
70.33
word_median
55
n_empty
0
n_duplicates
5
duplicate_rate
0.0004549
vocab_size
33,001
readability_flesch_mean
69.67
emoji_rate
0
url_rate
0.005095
one_word_rate
0.000182
allcaps_rate
0.0006368
boilerplate_rate
0.006459

location

text free_text
Short free-text place names averaging 2.98 words and 19.3 characters, with values like 'Prince Georges county', 'Cemetery', 'Cry Baby Bridge', and 'Wal-Mart' — likely the named site of a story or sighting (probably ghost/folklore lore given top entries). Vocabulary is dominated by 'school', 'cemetery', 'high', 'university', 'house', 'road', suggesting a strong tilt toward institutional and roadside locations. High cardinality (9903 unique of 10992) coexists with a 9.9% duplicate rate and one value ('Prince Georges county') hit 18 times, so most entries are unique strings but a long tail of repeated landmarks exists. Treatment: Normalise casing/whitespace and geocode or entity-link to extract structured place features before modelling. high · anthropic:claude-opus-4-7
n
10,992
nulls
3 (0.0%)
unique
9,903
len_min
2
len_max
1,016
len_mean
19.3
len_median
17
len_p95
34
word_mean
2.976
word_median
3
n_empty
0
n_duplicates
1,086
duplicate_rate
0.09883
vocab_size
8,232
readability_flesch_mean
45.79
emoji_rate
0
url_rate
0
one_word_rate
0.06334
allcaps_rate
0.003367
boilerplate_rate
0

state

categorical feature
This is a US state field with 51 distinct values (likely the 50 states plus DC) across 10,992 rows and no nulls. Distribution is broad with high entropy ratio (0.916); California leads at 9.73% (1,070 rows), followed by Texas (696) and Pennsylvania (649), roughly tracking population except Pennsylvania ranking unusually high above Florida (which doesn't appear in the top 10). Treatment: One-hot or target-encode for modelling; consider grouping low-frequency states. high · anthropic:claude-opus-4-7
n
10,992
nulls
0 (0.0%)
unique
51
top_value
California
top_rate
0.09734
cardinality
51
entropy
5.194
entropy_ratio
0.9157

state_abbrev

categorical feature
This is a US state abbreviation field with 51 distinct values (50 states plus presumably DC) and no nulls across 10,992 rows. Distribution is broadly spread (entropy ratio 0.916) with CA leading at 9.7% (1,070 rows), followed by TX (696) and PA (649), consistent with population-weighted geographic data. Treatment: one-hot or target-encode for modelling; safe to use as-is for grouping or joins. high · anthropic:claude-opus-4-7
n
10,992
nulls
0 (0.0%)
unique
51
top_value
CA
top_rate
0.09734
cardinality
51
entropy
5.194
entropy_ratio
0.9157

longitude

numeric feature
Geographic longitude coordinates, with 8774 unique values across 10992 rows and an 11.47% null rate. Values span -164.72 to 168.70 but cluster tightly between Q1 -99.12 and Q3 -80.30, indicating most records sit in North America despite the global range. Kurtosis of 15.2 confirms heavy tails from the 131 outliers (1.35%) reaching across the Pacific. Treatment: Pair with latitude for geospatial features; impute or filter the 11.47% nulls before mapping. high · anthropic:claude-opus-4-7
n
10,992
nulls
1,261 (11.5%)
unique
8,774
min
-164.7
max
168.7
mean
-92
median
-87.23
std
17.69
q1
-99.12
q3
-80.3
iqr
18.82
skew
0.2512
kurtosis
15.2
n_outliers
131
outlier_rate
0.01346
zero_rate
0

latitude

numeric feature
Geographic latitude in decimal degrees, ranging from -45.02 to 66.89 with a median of 39.28 and IQR of 7.20, suggesting most observations cluster in the northern mid-latitudes. The distribution is left-skewed (-0.97) with heavy tails (kurtosis 11.38) and 124 outliers (1.27%), consistent with a few southern-hemisphere points dragging the tail. Note the 11.47% null rate, which will need handling before any spatial analysis. Treatment: Pair with longitude for geospatial features; impute or drop the ~11% nulls before modelling. high · anthropic:claude-opus-4-7
n
10,992
nulls
1,261 (11.5%)
unique
8,775
min
-45.02
max
66.89
mean
38.34
median
39.28
std
5.259
q1
34.68
q3
41.87
iqr
7.197
skew
-0.971
kurtosis
11.38
n_outliers
124
outlier_rate
0.01274
zero_rate
0

city_longitude

numeric feature
Longitude of a city, with all 10992 values negative (min -164.72, max -67.84), placing every record in the Western Hemisphere and consistent with North American coverage. The distribution is left-skewed (skew -1.16) with median -87.09 pulled east of the mean -91.91, suggesting a concentration in the central/eastern US with a tail reaching toward Alaska or the Pacific. 5222 unique values across 10992 rows indicate many records share the same city, and 128 outliers (1.17%) likely correspond to far-western locations. Treatment: Use as a geospatial feature alongside latitude; consider clustering or binning rather than treating as a plain scalar. high · anthropic:claude-opus-4-7
n
10,992
nulls
29 (0.3%)
unique
5,222
min
-164.7
max
-67.84
mean
-91.91
median
-87.09
std
16.4
q1
-98.49
q3
-80.51
iqr
17.99
skew
-1.155
kurtosis
1.328
n_outliers
128
outlier_rate
0.01168
zero_rate
0

city_latitude

numeric feature
This column holds city latitudes in decimal degrees, ranging from 19.58 to 66.90 with a median of 39.28 and an IQR spanning 34.74 to 41.85. Values cluster firmly in the northern mid-latitudes (mean 38.38, std 5.07), with mild left skew (-0.32) and 128 outliers (1.17%) likely representing far-northern locales. Cardinality is moderate (5310 unique of 10992) and nulls are negligible (0.26%). Treatment: Pair with longitude for geospatial features; avoid treating as a standalone scalar. high · anthropic:claude-opus-4-7
n
10,992
nulls
29 (0.3%)
unique
5,310
min
19.58
max
66.9
mean
38.38
median
39.28
std
5.067
q1
34.74
q3
41.85
iqr
7.109
skew
-0.325
kurtosis
1.478
n_outliers
128
outlier_rate
0.01168
zero_rate
0

location_2

text feature allcaps
This column holds WKT geometry strings in the form `POINT(lon lat)`, with 8776 unique values across 10992 rows and an 11.47% null rate. Values are tightly bounded (length 22-43, always two 'words') and 9.81% are duplicates, with the most repeated coordinate appearing 20 times — suggesting clusters at recurring sites. The 'allcaps' alert is a false positive driven by the literal `POINT` prefix rather than natural language. Treatment: Parse into numeric longitude/latitude pair before any geospatial modelling. high · anthropic:claude-opus-4-7
n
10,992
nulls
1,261 (11.5%)
unique
8,776
len_min
22
len_max
43
len_mean
30.81
len_median
29
len_p95
36
word_mean
2
word_median
2
n_empty
0
n_duplicates
955
duplicate_rate
0.09814
vocab_size
17,549
readability_flesch_mean
120.2
emoji_rate
0
url_rate
0
one_word_rate
0
allcaps_rate
1
boilerplate_rate
0

city_location

text feature allcaps duplicates
Despite the name, city_location stores WKT POINT(longitude latitude) geometry strings, not city names — every value matches that format (allcaps_rate 1.0, word_mean 2.0). Roughly half the rows are repeats (duplicate_rate 0.516, n_duplicates 5652) with the top coordinate (Los Angeles, 34.05/-118.24) appearing 61 times, suggesting a finite set of city centroids reused across records. Cardinality is still high (5311 unique of 10992) and nulls are negligible (0.26%). Treatment: Parse the POINT() strings into numeric lat/long pairs before any geospatial modelling. high · anthropic:claude-opus-4-7
n
10,992
nulls
29 (0.3%)
unique
5,311
len_min
22
len_max
43
len_mean
31.19
len_median
29
len_p95
36
word_mean
2
word_median
2
n_empty
0
n_duplicates
5,652
duplicate_rate
0.5156
vocab_size
10,532
readability_flesch_mean
120.2
emoji_rate
0
url_rate
0
one_word_rate
0
allcaps_rate
1
boilerplate_rate
0