Correction to ‘Personality composition is more important than group size in determining collective foraging behaviour in the wild’

Proc. R. Soc. B 281, 20141424 (Published online 7 December 2014) (doi:10.1098/rspb.2014.1424)

The first author of ‘Personality composition is more important than group size in determining collective foraging behaviour in the wild’ (https://doi.org/10.1098/rspb.2014.1424) recently discovered a series of errors in the raw data associated with this manuscript. In February 2014, these data were collected jointly by Keiser and Pruitt who recorded behavioural data daily into shared notebooks. One of these notebooks has been retained by Keiser, from which the majority of the data can be cross-validated. Working with a group of editors at Proceedings B, we discovered the following errors in the dataset used for analyses and subsequently uploaded to the Dryad Digital Repository [1]:

1. Autofill errors in excel sheet available on the Dryad Digital Repository

Colony #51 contained 30 spiders, present on row 83 on the Dryad Digital Repository data sheet. Instead of copying the number 30 from row 83 through row 85, excel auto-filled the numbers 31 and 32. Similarly, there was an autofill error in the ‘source colony ID’ column, because Colony #51 and #52 were both collected from Source colony N10-A4. On the Dryad Digital Repository datasheet, rows 84 and 85 depict Colony #51 coming from three different source colonies (N10-A4, N10-A5 and N10-A6). After checking the original data notebook, there are no N10-A5 and N10-A6. These autofill errors have been corrected in an updated data sheet available on Figshare (https://doi.org/10.6084/m9.figshare.11861400.v1).

2. Miscoding of independent variables in four rows

On lines 218 and 224 in the Dryad Digital Repository excel file, the independent variables (colony size and group composition) were reciprocally mislabelled between Replicate #39 and #41 (i.e. line 218 should be Replicate 41 and line 224 should be Replicate 39). This was verified from the original data notebook. A similar error occurred between lines 215 and 221 (i.e. line 215 should be Replicate #40 and line 221 should be Replicate #38). The miscoding of independent variables has been corrected in the new data sheet available on Figshare (https://doi.org/10.6084/m9.figshare.11861400.v1).

3. Duplication of rows during data entry

There were duplication events in entering data from notebooks to the excel file, and subsequently, the JMP file used for analyses. Rows 19 and 20, 24 and 25, 29 and 30, 47 and 48, and 49 and 50 represent cases where lines were entered identically back-to-back. We cross-validated with original data notebooks to verify that the values in these lines were entered correctly, but accidentally entered twice. The following rows contain identical values, but were not the product of data entry errors: 26, 28, 37, 39, 73, 83. These data match exactly as they were recorded in the field notebook, and as such have not been removed prior to re-analyses. There were additional rows containing collective foraging and/or collective anti-predator responses that contain identical values, but these data were not recorded in the notebook retained by Keiser and cannot be cross-validated. We have taken a conservative approach and removed all of these identical values prior to reanalysis (rows 146, 149, 159, 163, 164, 184, 209, 214, 215, 244, 276).

4. Clarification of methods

Our methods stated ‘We first tested each colony three times before they produced a capture web’. However, 11 out of 64 colonies were only assayed twice ‘without a capture web’ before they were placed in trees outside, while the remaining colonies were assayed three times. We are uncertain as to the circumstances that led to these colonies receiving one less assay than the rest.

Our methods stated ‘The foraging behaviour of each colony was measured between three and 16 times in total’. We did not sufficiently explain the reason for this variability. After building a capture web, the foraging behaviour of each colony was measured as many times as possible before they succumbed to attacks by predatory ants. The number of hours until each colony was raided by ants is now included in the dataset on Figshare, and these data were included in a publication elsewhere [2].

In our methods section, we failed to describe that between 8 and 12 spiders were observed simultaneously during individual boldness assays. Spiders were placed individually into their own arenas, and each arena had a separate stopwatch assigned to it. Arenas were arranged in a half circle around each observer so interacting with one spider did not cast shadows over or disturb neighbouring spiders. Our field site did not have air conditioning, so spiders assayed at different times of the day were subject to natural variation in temperature during daylight hours.

5. Updated analyses

After correcting/removing all errors from the data sheet, Keiser reanalysed the web size and collective foraging data using models identical to those in the original manuscript. We now compare p-values to Bonferroni-adjusted α values (0.05/3 = 0.017) to correct for multiple testing. In summary: though the p-values, degrees of freedom and test statistics have changed, all of the conclusions garnered from the original study remain entirely supported (tables 1 and 2).

Table 1.

Summary of independent variables predicting web size. Statistically significant effects are bolded.

source	d.f.	d.f.den	F	p-value
web size
size	1	40.359776738	33.6337	<0.0001
composition	3	37.698588662	0.8617	0.4694
composition × size	3	38.009250682	0.8061	0.4983

Table 2.

Summary of independent variables predicting three factors of colonies’ collective foraging behaviour. Statistically significant effects are bolded.

source	d.f.	d.f.den	F	p-value
latency to emerge
size	1	246.51205788	0.9193	0.3386
composition	3	243.68252303	5.0338	0.0021
web presence	1	240.77671377	0.1182	0.7313
composition × size	3	244.84562091	0.5437	0.6528
web presence × size	1	240.62834824	1.2248	0.2695
web presence × composition	3	240.076234	5.5140	0.0011
composition × size × web presence	3	240.97618345	0.8657	0.4595
latency to attack
size	1	244.67026889	2.4437	0.1193
composition	3	241.83841202	5.3581	0.0014
web presence	1	238.70868433	1.4825	0.2246
composition × size	3	242.94381849	0.2121	0.8880
web presence × size	1	238.59712335	1.4651	0.2273
web presence × composition	3	238.11763864	7.5607	<0.0001
composition × size × web presence	3	238.87697887	2.4537	0.0639
number of attackers
size	1	239.89311211	27.2822	<0.0001
composition	3	241.07231586	14.2091	<0.0001
web presence	1	238.78719339	11.5898	0.0008
composition × size	3	239.02618659	3.9964	0.0084
web presence × size	1	236.29699761	0.0007	0.9784
web presence × composition	3	237.72696872	0.4651	0.7069
composition × size × web presence	3	237.59542258	2.6271	0.0510

The web size data were analysed with a general linear mixed-model with the following predictor variables: group size, colony composition and a group size × colony composition interaction term. We also included source colony ID as a random effect. Web size was predicted only by group size; group composition had no detectable influence on total web area (table 1).

The latency to emerge and latency to attack data were log-transformed to meet model assumptions. We used general linear mixed models to predict three response variables (latency to emerge, latency to attack, number of attackers) each with the following predictor variables: group size, colony composition, web presence (with/without capture web), group size × colony composition, group size × web presence, composition × web presence and group size × colony composition × web presence. Source colony ID was included as a random effect. Only group composition and the interaction between group composition and web presence predicted colonies' latency to emerge and latency to attack (table 2). The number of attackers was influenced jointly by group size and group composition (table 2).

6. Conclusion

Although the authors found errors in the dataset used for analyses, we were able to verify which data were faulty and which were reliable by validating with original data notebooks. Thus, the authors stand by the conclusions garnered from the original manuscript.