Why do Android app users make relatively so many more casual observations than other users?

General
1

Just out of curiosity, I looked at some statistics for observations created via iOS app vs those created via Android app vs all observations. I was originally thinking that far fewer birds would be observed via app, and I was hoping to see that in the statistics (I did), but I noticed something more interesting…

Across all 210 million observations, a little more than 21.2 million (11%) are casual grade, and of those, roughly 20% of casual grade fall into the “unknown” iconic taxon group (not identified, life, bacteria, viruses, etc.).

For 65 million observations created via iOS (30% of all observations), a bit more than 6.8 million (11%) are casual grade, and roughly 12% of casual grade fall into “unknown”.

For 54 million observations created via Android (25% of all observations), roughly 8.8 million (16%) are casual grade, and of those, 40% are “unknown”.

Looking at the raw number of casual grade “unknowns”, there are more than 4.2 million in total, with 0.8 million (~20%) created via iOS and another 3.5 million (~83%) created via Android. Digging further into the 3.5 million Android-created casual “unknowns”, 2.6 million of these (75%) either have location missing (probably most of these) or have location privacy set to private.

So to summarize:

Source Total Obs Casual Obs Casual “Unknown”
All 210 million 21.2 million 4.2 million
iOS 65 million 6.8 million 0.8 million
Android 54 million 8.8 million 3.5 million

I was expecting proportionally more casual grade observations to be created via app (since I assume the apps are used more by casual users), and that seems to be true, but it’s especially true for observations created via Android app. I’ve never used the iOS app, but I can’t think of anything about the design of the Android app specifically that would make it so much more likely to generate casual “unknown” observations (especially ones without location).

So then is there something about Android app users?

I don’t think it’s due to relative inexperience with the apps. The summary below doesn’t seem to show Android users making fewer observations than iOS users:

Source Observations Observers Obs / Observers
All 212 million 4.0 million ~536
iOS 65 million 2.1 million ~31
Android 54 million 1.5 million ~37

I know that observations created via iOS are relatively high in the USA, UK, Scandinavia, and East Asia, and I know that observation created via Android are relatively high in Europe, Mexico, South America, South and Southeast Asia, and West Africa.

Maybe excluding Europe, it would seem like maybe folks creating observations via Android would likely be younger people (although I’m not sure how to verify that). So is high number of Android casuals due to younger users? Or is there something more to it?

What do you think?

1 Like
2

Does the iPhone has the ability to switch off the gps for the telephone? Can’t find it yet.
https://support.apple.com/nl-nl/102647

The camera of the iPhone is better, makes better pictures with identifiable organisms which results in more RG observations. Cheap android phones with bad camera’s makes photo’s with unidentifiable organisms which results in more NeedsID observations.

Maybe add the percentage in the to the summerizing table… or start the text with the table…

2 Likes
3

I’ve had times where the Android app fails to gather location data from photos I’ve taken, even if I definitely had GPS and location tagging on. There was a period where there was a bug with how the app interacted with Google Photos that might have caused at least a few no-location observations, for example. It could be that these issues cause a lot of observations with missing dates or locations.

optilete’s note about location being turned off is another good point; if people are turning off location tracking and then taking their photos, that won’t go very well for generating a location. If iOS users just can’t disable location tracking, that would eliminate another possible cause of data-quality-casualness.

Of course, “casual” really conflates two things: data quality issues, and captive/cultivated organisms. It would be interesting if we could distinguish between those and see if the two cases play out differently. Maybe Android users are more likely to, for example, be students recruited into using iNaturalist by educators, who end up targeting cultivated plants near their classroom; or maybe regions where Android is more prevalent as a platform also have different urban conditions or other factors that lead to more observation of non-wild organisms. Though that feels more speculative than known technical limitations.

2 Likes
4

How well are the apps translated into languages other than English? Your list of where relatively more users are seems to suggest that Android users are skewed towards more non English-speaking populations, so maybe language barrier issues are a contributing factor to the trends you observed.

1 Like
5

I think to get to the bottom of this (if it is even a real difference), we would need to know why those records are marked casual. Is it a location problem, or is there something else.
Might be interesting to just compare records from the US to eliminate differences due to phone access, etc.

1 Like
6

i don’t think this is possible, since if you don’t have a location on the observation, you can’t really tell where the observation came from.

i think it’s a location problem. i mentioned above that 2.6 million of 3.5 million Android-generated casual “unknowns” were either missing location or had geoprivacy set to private. so if i back out those 2.6 million records from the Android numbers, I get 51.4 remaining total observations, with 6.2 million remaining casuals (12%), and 0.9 million remaining casual “unknowns” (15% of remaining casuals), which puts things much more in line with iOS and all observations.

7

I guess this isn’t necessarily relevant but I follow a couple of my former professors who regularly use Seek and have it set to automatically upload observations to iNaturalist. Most of their observations end up Casual for one reason or another because some of the metadata doesn’t get imported properly (missing location, date, photo, or even blank Unknown observations…). Could well be something similar going on at a smaller scale with the normal apps. Are you able to compare the stats for observations coming from Seek?

8

There was an Android bug a few years ago that would sometimes split observations when uploaded - one observation with just a photo, one with just the data, so both were casual. I suspect that’s a factor.

3 Likes
9

okay. maybe that’s it. do see a period between 2018 through maybe 2023 where it does seem like among casual “uknowns”, the set of geo=false was particularly high:

Created Year Missing Loc or Private Total %
2012 55 183 30.1%
2013 442 1,060 41.7%
2014 601 1,074 56.0%
2015 1,287 2,577 49.9%
2016 11,128 16,886 65.9%
2017 18,136 26,451 68.6%
2018 174,846 203,692 85.8%
2019 305,948 370,769 82.5%
2020 372,371 445,686 83.6%
2021 443,989 543,507 81.7%
2022 497,159 635,368 78.2%
2023 481,915 658,932 73.1%
2024 296,912 598,234 49.6%
2 Likes
10

these are the stats from observations from Seek: https://jumear.github.io/stirfry/iNat_obs_counts_by_iconic_taxa?oauth_application_id=333.

the percent of casual is relatively high vs all observations, but this is true across all iconic taxon groups (particularly plants and mammals). so i suspect that means Seek users are recording a lot of captives or maybe those observations are relatively hard to identify to species.

11

I would also wonder if CNC has some part in this. I would guess that new/duress users are least likely to realize when there are problems with observations, know how to fix them, or be motivated to find out how/respond to info on how. It would be interesting to see the monthly breakdown of these observations and if there is a spike in April/May with CNC compared to other months.

12

I bet there’s a spike around the CNC for everything, whether good or bad.

2 Likes
13

For absolute numbers, absolutely, but proportionally there might be more of these types of observations during CNC, as @pisum looked at proportions of these observations by year above which accounts for the increase in absolute number of observations across years. And CNC observations have increased yearly (though I am not sure whether this is in line with/above/below overall increases in observations on iNat).

14

what i see since 2018 (below) is that there definitely are more casuals in CNC months, but it doesn’t seem that bad to me. there definitely are other months like Feb 2019 that stand out more to me as having unusually high numbers of casuals.

one thing that i do see is that the casual ratio is lower in more recent months. i’m not sure if that’s a function of older observations having had a longer time to get identifiers looking at them and classifying them properly (or, alternatively, more observations in absolute terms in recent months having overwhelmed the identifier community) or if there is a genuine improvement in more recent years.

one really interesting thing (to me) in the monthly numbers to me is starting around this year, Android seems to have edged out iOS as the largest source of observations. (that’s especially interesting when you consider my previous note about the geographic regions that have high numbers of Android users.)

Created Month All Total All Casual All Casual / Total Android Total Android Casual Android Casual / Total iOS iOS Casual iOS Casual / Total
1/1/2018 337,944 44,053 0.130 31,175 5,884 0.189 68,083 7,835 0.115
2/1/2018 332,136 48,572 0.146 39,564 8,317 0.210 80,627 9,909 0.123
3/1/2018 471,622 57,655 0.122 81,699 19,024 0.233 147,804 20,465 0.138
4/1/2018 1,073,190 173,598 0.162 227,360 56,425 0.248 443,395 77,033 0.174
5/1/2018 988,261 138,763 0.140 205,995 55,967 0.272 381,132 58,898 0.155
6/1/2018 871,475 103,155 0.118 188,603 47,271 0.251 321,634 36,174 0.112
7/1/2018 998,591 165,286 0.166 195,393 50,347 0.258 331,143 33,915 0.102
8/1/2018 1,019,235 216,523 0.212 181,934 46,463 0.255 313,268 33,800 0.108
9/1/2018 912,432 136,622 0.150 197,798 57,505 0.291 353,370 51,354 0.145
10/1/2018 841,344 115,575 0.137 174,748 53,476 0.306 307,362 41,694 0.136
11/1/2018 704,667 140,743 0.200 135,955 51,220 0.377 234,603 61,659 0.263
12/1/2018 573,141 57,951 0.101 84,335 22,147 0.263 143,638 19,407 0.135
1/1/2019 626,140 62,696 0.100 90,612 22,894 0.253 148,690 21,502 0.145
2/1/2019 765,663 200,042 0.261 190,755 97,670 0.512 182,675 41,304 0.226
3/1/2019 954,559 118,378 0.124 191,498 47,275 0.247 318,930 45,094 0.141
4/1/2019 2,272,726 422,623 0.186 647,063 190,698 0.295 859,914 136,227 0.158
5/1/2019 1,828,781 251,024 0.137 427,271 106,108 0.248 667,261 91,170 0.137
6/1/2019 1,738,770 196,067 0.113 425,690 91,605 0.215 652,339 74,729 0.115
7/1/2019 1,894,877 195,747 0.103 462,848 90,896 0.196 674,631 65,764 0.097
8/1/2019 1,858,526 203,110 0.109 450,487 91,789 0.204 649,730 66,158 0.102
9/1/2019 1,712,732 213,486 0.125 400,197 86,143 0.215 652,348 85,581 0.131
10/1/2019 1,405,707 182,071 0.130 334,242 77,602 0.232 475,259 65,360 0.138
11/1/2019 1,036,851 150,821 0.145 233,946 64,841 0.277 291,807 56,493 0.194
12/1/2019 874,135 90,539 0.104 172,277 38,563 0.224 213,151 30,646 0.144
1/1/2020 976,079 95,882 0.098 174,036 34,150 0.196 235,833 28,788 0.122
2/1/2020 1,173,542 252,193 0.215 204,325 41,055 0.201 261,456 34,347 0.131
3/1/2020 1,681,279 268,452 0.160 323,869 63,398 0.196 497,240 63,502 0.128
4/1/2020 3,144,957 436,080 0.139 745,845 163,649 0.219 1,132,316 179,022 0.158
5/1/2020 3,477,538 413,428 0.119 887,022 165,323 0.186 1,249,808 165,301 0.132
6/1/2020 3,315,133 351,014 0.106 890,885 154,515 0.173 1,129,997 122,992 0.109
7/1/2020 3,301,743 311,407 0.094 889,214 142,069 0.160 1,111,657 102,554 0.092
8/1/2020 3,082,797 285,460 0.093 809,159 132,628 0.164 1,017,505 97,035 0.095
9/1/2020 3,015,813 342,643 0.114 750,734 130,007 0.173 1,069,087 133,667 0.125
10/1/2020 2,465,066 289,207 0.117 592,692 112,811 0.190 805,399 115,655 0.144
11/1/2020 1,902,416 298,875 0.157 446,644 100,077 0.224 553,446 125,212 0.226
12/1/2020 1,409,062 151,890 0.108 295,785 55,992 0.189 316,008 44,439 0.141
1/1/2021 1,453,469 131,028 0.090 311,642 60,615 0.195 320,302 37,225 0.116
2/1/2021 1,409,831 149,322 0.106 293,920 51,729 0.176 338,210 49,135 0.145
3/1/2021 2,118,900 236,446 0.112 501,676 92,756 0.185 619,985 86,945 0.140
4/1/2021 3,262,524 376,008 0.115 836,557 147,254 0.176 1,121,053 146,532 0.131
5/1/2021 4,960,055 580,448 0.117 1,429,016 255,285 0.179 1,581,501 188,227 0.119
6/1/2021 3,889,463 378,217 0.097 1,097,442 166,139 0.151 1,250,259 111,966 0.090
7/1/2021 4,113,958 373,524 0.091 1,141,064 155,448 0.136 1,274,272 92,619 0.073
8/1/2021 3,675,674 310,387 0.084 985,579 133,837 0.136 1,111,289 85,532 0.077
9/1/2021 3,406,093 330,507 0.097 894,532 130,588 0.146 1,133,554 120,485 0.106
10/1/2021 3,074,436 287,925 0.094 795,056 126,712 0.159 959,322 99,435 0.104
11/1/2021 2,254,284 368,541 0.163 533,222 112,216 0.210 636,788 110,904 0.174
12/1/2021 1,727,492 157,404 0.091 383,270 63,220 0.165 423,244 46,726 0.110
1/1/2022 1,796,634 147,796 0.082 368,869 56,191 0.152 421,042 42,106 0.100
2/1/2022 1,670,226 139,998 0.084 366,374 56,498 0.154 416,020 49,893 0.120
3/1/2022 2,298,880 209,204 0.091 553,754 88,367 0.160 711,049 80,748 0.114
4/1/2022 3,840,129 411,336 0.107 1,101,144 188,005 0.171 1,337,933 155,682 0.116
5/1/2022 5,457,339 620,655 0.114 1,627,412 295,649 0.182 1,716,573 187,423 0.109
6/1/2022 4,370,746 346,264 0.079 1,273,600 174,887 0.137 1,412,447 104,365 0.074
7/1/2022 4,579,764 311,195 0.068 1,299,491 156,287 0.120 1,464,126 90,862 0.062
8/1/2022 4,109,297 329,791 0.080 1,109,781 140,174 0.126 1,246,457 85,633 0.069
9/1/2022 3,822,817 323,558 0.085 1,011,120 142,086 0.141 1,284,681 121,353 0.094
10/1/2022 3,511,831 315,905 0.090 985,657 143,413 0.145 1,060,194 102,248 0.096
11/1/2022 2,582,145 280,208 0.109 726,950 144,952 0.199 686,584 85,826 0.125
12/1/2022 2,086,915 174,300 0.084 473,761 63,778 0.135 507,679 49,303 0.097
1/1/2023 2,196,681 140,611 0.064 505,312 68,705 0.136 530,776 40,297 0.076
2/1/2023 2,082,034 145,483 0.070 491,415 66,644 0.136 544,590 48,332 0.089
3/1/2023 2,952,048 227,082 0.077 746,711 101,103 0.135 871,047 78,397 0.090
4/1/2023 5,236,757 496,784 0.095 1,648,303 252,323 0.153 1,737,730 170,645 0.098
5/1/2023 6,320,331 587,798 0.093 1,954,254 302,255 0.155 1,970,334 176,546 0.090
6/1/2023 5,306,521 358,487 0.068 1,600,577 187,190 0.117 1,695,649 103,622 0.061
7/1/2023 5,522,359 310,440 0.056 1,645,703 160,405 0.097 1,725,819 87,240 0.051
8/1/2023 4,975,272 299,388 0.060 1,464,189 152,399 0.104 1,506,578 80,692 0.054
9/1/2023 4,401,112 315,609 0.072 1,177,843 124,867 0.106 1,409,815 99,960 0.071
10/1/2023 3,858,025 240,575 0.062 1,026,927 106,906 0.104 1,193,528 85,095 0.071
11/1/2023 2,987,732 226,169 0.076 798,151 95,544 0.120 818,801 72,239 0.088
12/1/2023 2,676,743 178,805 0.067 670,708 85,970 0.128 669,948 57,631 0.086
1/1/2024 2,607,488 192,799 0.074 598,343 62,509 0.104 630,822 45,629 0.072
2/1/2024 2,699,785 190,077 0.070 695,055 87,523 0.126 693,481 55,738 0.080
3/1/2024 3,885,323 279,878 0.072 1,086,510 137,978 0.127 1,215,833 94,780 0.078
4/1/2024 7,412,013 797,288 0.108 2,626,290 454,746 0.173 2,410,338 220,985 0.092
5/1/2024 6,089,869 455,535 0.075 1,948,384 245,189 0.126 1,893,336 131,786 0.070
1 Like
15

Interesting. There’s a little bit of a lift in CNC months, but not dramatic. November actually seems to be consistently higher than others - wonder what might be driving that?

16

my first thought is the Great Southern Bioblitz, but i didn’t verify.

… or maybe that’s a good month to do a bioblitz after school starts in the fall?

1 Like
17

I hope for a GSB spike, but on iNat’s annual graphics it is still a ‘little’ spike compared to the CNC SPIKE!

1 Like
18

I have often wondered what it is about my Android phone that makes it preferentially focus on the background instead of the subject. I can position it so that the subject is prominent and dead-center, and still it blurs the subject and focuses crisply on the background. I actually yell at my phone often.

Even more aggravating, often it will show me a nice, crisp image until I click the shutter, but the resulting photo is blurrier than the preview. Why?

3 Likes
19

it sounds like 3 possible things:

  • your subject is small and does not fill much of of the area that your phone is scanning to focus. in this case, you either need to focus on a proxy at the distance, or else you need an app that allows you to use manual focus.
  • you might be trying to focus right on the edge of the minimum focusing distance of your phone. even if you are beyond the minimum distance, if you tap to take the photo, you could just slightly push your phone across that border, causing the phone to lose focus. (or if there is wind or you have a shaky hand, your subject could also be moved closer than your phone is able to focus properly.)
  • it’s possible that your phone’s subject tracking algorithm isn’t very robust. so it will just lose the intended subject easily.
2 Likes