Evidence for a Two Component Cognitive Structure in Pigeons (Columba livia)

doi:10.21203/rs.3.rs-3295565/v1

Download PDF

Research Article

Evidence for a Two Component Cognitive Structure in Pigeons (Columba livia)

https://doi.org/10.21203/rs.3.rs-3295565/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 02 Nov, 2024

Read the published version in Animal Cognition →

You are reading this latest preprint version

A well replicated result in humans is that performance, whether good or bad, is consistent across a wide variety of cognitive tasks. Factor analysis consistently extracts one factor that can account for approximately half of the variance in performance. This factor is termed g and almost all cognitive tasks positively load onto this factor. While some neurobiological correlates of g have been identified in humans, causal experiments are only feasible in animals. When mice and some avian species are assessed with cognitive test batteries, performance positively correlates, and the first component extracted has similar properties to g. There are some limitations to the species tested thus far, including comparability in the cognitive domains assessed across species and homogeneous samples. The pigeon is an ideal subject to overcome these issues since pigeons, humans, and other primates are frequently given similar tasks and many neural correlates of performance have been identified in the pigeon. We created a test battery that assessed different domains, including associative learning, memory, cognitive flexibility, and reaction time. Yet we did not consistently extract a g like factor. Analyses indicated a two-component structure with differential task loadings. Possible interpretations of the components are associative learning/memorization versus a general rule, degree of automaticity, and sensitivity to age related decline. Reasons and implications for this two-component structure are discussed.

intelligence

test battery

pigeon

g factor

Does performance on a vocabulary test give any meaningful indication on how someone will perform on a math test? Or how quickly they react to a change in an array of stimuli? Or how well they can mentally rotate an image? Surprisingly, the answer is yes. Performance across all these tasks shows a positive correlation; if a person performs well on one task, they are likely to perform well in another. This effect has been replicated many times, but the most compelling results are from full scale intelligence quotient (FSIQ) tests because of the number and variety of tasks used. The exact number and type included vary from test to test, but they typically include 11-17 measures that assess memory, basic math, spatial reasoning, and analogical reasoning (Carroll, 1993; Johnson et al., 2004). Despite differences in the test batteries and variety of tasks used, a positive correlation matrix is found (Carroll, 1993; Johnson et al., 2004). When variable-reducing techniques, like principal component analysis (PCA) or factor analysis, are applied to this positive correlation matrix, one factor is consistently extracted that can account for approximately half of the variance (Carroll, 1993; Deary, 2000). All cognitive tasks positively load onto this factor, meaning the factor can account for variance in performance in the task (Carroll, 1993; Deary, 2000). Because this factor is seemingly related to all cognitive abilities, it is referred to as g for “general” (Spearman, 1904).

g has been extracted with a variety of test batteries in a variety of samples, making it one of the most well-replicated results in psychology (Carroll, 1993; Deary, 2000; Johnson et al., 2004; Warne & Burningham, 2019). Despite the ubiquity of g, there are still important parameters to consider when creating a test or test battery to extract this factor. As mentioned earlier, almost all cognitive tasks load onto g, but some tasks have a higher loading than others. The highest loadings on the g factor will be found when tasks are complex, novel, and require reasoning, irrespective of the task content or method of delivery (Jensen, 1992; Quiroga et al., 2019). Even though we know what kinds of tasks load highly onto g, no task is a perfect or pure measure of any specific cognitive construct. All cognitive tasks assess g to some degree, but they also assess more specific abilities (Gignac, 2015). Using a large and diverse battery of tests will attenuate task-specific variance, resulting in a more accurate g factor (Major et al., 2011). Another issue with all measures is random error, variables unrelated to the construct of interest that impact performance on the task. Random error can cause performance to be different when participants complete the measure at different time points or respond differently to theoretically similar items. If a measure produces similar results, despite random error, it is referred to as a reliable measure (John & Benet-Martinez, 2000). Reliable measures are crucial because more of the variance in performance across individuals is due to differences in the actual cognitive ability the task is measuring instead of differences caused by random error (Bray et al., 1998). Reliability also impacts the correlation matrix. Less reliable measures will artificially lower the correlations, which impacts the extraction of the g factor (John & Benet-Martinez, 2000). While it is important that measures are reliable, they also need to be sensitive enough to detect individual differences across people. The g factor accounts for variance in performance across people, therefore the tasks used should show variability based on true differences in cognitive ability (Hedge et al., 2018). Extracting a robust g factor depends on using appropriate tasks and on the sample that is being assessed. Highly homogeneous samples of human participants may not have true differences in the construct of interest which reduces variance and attenuates the subsequent g factor (Sackett & Yang, 2000). It is also best to test many participants due to how correlational and factor analyses are conducted. While a sufficient sample size to detect a reliable correlation depends on a variety of parameters, the sample size required can be in the hundreds (Bonett & Wright, 2000), and for factor analysis a sufficient sample size ranges from as few as 75 to as many as 1,200 participants (Mundfrom et al., 2005, but see de Winter et al., 2009). To summarize, to extract the strongest g factor, a large, heterogeneous sample of people should be given a large variety of cognitive tasks, that are reliable and sensitive to individual differences.

Even though g is consistently replicated, it is still not clear what exactly g is. It is tempting to use intelligence and g interchangeably, since g can be extracted using FSIQ tests and is related to a wide variety of cognitive tasks. Yet g only accounts for half of the variance in performance on FSIQ, which means other factors besides g are related to performance. In addition, the amount of variance g can account for relies on the strength of the correlation matrix (Jensen, 1998). Individuals who perform better on intelligence tests tend to have weaker correlations between tasks, meaning that higher FSIQ scores come with increased differentiation of abilities (Blum & Holling, 2017). For higher performing individuals, g explains less of the variance in performance (Jensen, 1998). Therefore, it would be incorrect to say that more intelligent people have more g (Detterman, 1991). These results indicate that when we refer to intelligence in a colloquial sense, we are referring to more than g, even though the two concepts are closely related (Jung & Haier, 2007; Stankov, 2017).

With that distinction stated, understanding g is still important given the consistent pattern of correlations across tasks, even among high ability individuals (Blum & Holling, 2017). Even though g is a single factor or component, it does not mean it is a single causal entity. Instead, g is commonly theorized to be composed of more specific cognitive processes like working memory (WM), short term memory (STM), processing speed, attention, and associative learning (Conway et al., 2002; Deary, 2000; Jensen, 1998; Kaufman et al., 2009; Sheppard & Vernon, 2008). It is likely that the tasks included in FSIQ tests, particularly complex tasks that load highly onto g, require support from multiple cognitive domains (Chuderski, 2013). Therefore, g could reflect individual differences in how many processes are required to solve a task (Chuderski, 2013). Another theory suggests that differences in one of these abilities could act as a bottleneck, constraining and weakening the ability of all other cognitive domains to function (Kovacs & Conway, 2019). With this theory, g primarily reflects differences in one cognitive ability, but it is unclear which cognitive ability. These theories are helpful for understanding the more specific cognitive processes that are involved with intelligence tests and how those processes are used across a large number of tasks (Conway et al., 2002; Deary, 2000; Jensen, 1998; Kaufman et al., 2009; Sheppard & Vernon, 2008). Future research is still needed, however, to fully understand if there is a relationship between these cognitive processes that could impact the positive correlation matrix (Frischkorn et al., 2019).

At the psychological construct level, g is related to a variety of cognitive processes. Similarly, g and intelligence are correlated with a variety of neurobiological mechanisms, processes, and features, (Deary et al., 2010). Thus far there have been two major lines of research. One focuses on what makes individuals different, for example comparing people who have high IQ scores to people who do not. The most robust result from this line of research is the modest positive correlation between brain size and measures of intelligence (Pietschnig et al., 2015). The other line of research focuses on why performance is positively correlated across tasks, irrespective of individual performance. A variety of methods, including functional magnetic resonance imaging (fMRI), positron emission tomography (PET), and lesions due to accident or stroke, have indicated the importance of the frontal cortex in a wide variety of tasks (Jung & Haier, 2007). The dorsolateral prefrontal cortex in particular is active during a variety of WM and reasoning tasks, though similar patterns of activations in different areas of the frontal cortex for other types of tasks have also been identified (Colom et al., 2013). Yet, brain areas do not function in isolation; rather, different areas are connected, forming functional networks (van den Heuvel & Sporns, 2017). Similar performance across tasks may be partially due to how whole networks are activated by tasks instead of discrete regions. A network connecting the frontal and parietal cortex is implicated (Fraenz et al., 2021; Jung & Haier, 2007; Zanto & Gazzaley, 2013). Thus, most research has indicated that consistent performance could be due to activation of the frontal cortex over a wide variety of tasks (Colom et al., 2013; Jung & Haier, 2007)

Important neurobiological correlates of g and intelligence have been identified in humans, but the techniques used thus far fail to support causal interpretations. Nonhuman animal models (hereafter animals) would be ideal to explore causal manipulations, but it first needs to be established that animals have a g factor similar to what is seen in humans (Matzel et al., 2013). Investigations over the past 20 years have generated promising results that are described in more detail elsewhere (Burkart et al., 2017; Flaim & Blaisdell, 2020; Shaw & Schmelz, 2017), but some key results from mice and avian species will be briefly reviewed here. For mice, Matzel and colleagues in particular have been consistently exploring a general factor using a cognitive test battery that targets different domains of learning (Matzel et al., 2003; but also see Locurto & Scanlon, 1998; Locurto et al., 2003, 2006). Briefly, the test battery includes five tasks and measures non-spatial navigation (Lashley III maze), spatial navigation (Morris water maze), suppression of exploratory behavior to avoid an aversive audiovisual stimulus (passive avoidance), using odor to guide a response (odor discrimination), and using an auditory cue to predict an aversive shock (associative fear learning). Multiple experiments found that performance was positively correlated across all tasks, and the first factor extracted could account for 38-43% of the variance in performance (Kolata et al., 2005, 2007; Matzel et al., 2003; 2006). For these individual experiments however, the number of subjects ranged from 21-56, which is smaller than what is typically used or recommended in human studies (Mundfrom et al., 2005, but see de Winter et al., 2009). When the results from multiple experiments were combined to have a total of 241 subjects, the result was replicated, providing robust evidence for a g like factor in mice (Kolata et al., 2008). Subsequent experiments have shown that performance on this cognitive test battery is positively correlated with measures of WM, similar to what is seen in humans (Kolata et al., 2005). Investigations with avian species have also yielded interesting results. Cognitive test batteries that typically include motor learning, color discrimination, reversal learning, spatial memory, and inhibitory control, have been administered to robins (Shaw et al., 2015), spotted bowerbirds (Isden et al., 2013), magpies (Ashton et al., 2018), southern pied babblers (Soravia et al., 2022), and song sparrows (Anderson et al., 2017; Boogert et al., 2011). For robins and spotted bowerbirds, performance across the tasks was mostly positively, though not significantly, correlated, and a factor that could explain 34% and 44% of the variance in performance, respectively, was found (Isden et al., 2013; Shaw et al., 2015). This result should be treated with some caution since a small number of subjects, 16 robins and 14 bowerbirds, were assessed. More robust results have been obtained with magpies, which assessed 56 subjects (Ashton et al., 2018) and southern pied babblers, which assessed 38 subjects (Soravia et al., 2022). The subsequent correlation matrices were uniformly positive and the subsequent factor extracted accounted for 64% and 60% of the variance respectively (Ashton et al., 2018; Soravia et al., 2022). Yet, similar results were not found in song sparrows, even though 52 (Boogert et al., 2011) and 41 (Anderson et al., 2017) birds were assessed using the same test battery. Across both experiments, two factors were extracted, and not all tasks loaded onto the first factor extracted. This may be due to the low reliability in performance across years on cognitive tasks in song sparrows (Soha et al., 2019). While the results from animals thus far are interesting and promising, there are some difficulties in comparing g across species.

Research with many species thus far indicates that g can be found beyond humans, but it is not clear exactly how similar g is across species. This ambiguity is partially due to the differences in test batteries across species. In humans, g has been heavily investigated in relationship to processing speed, where more intelligent individuals are consistently faster on simple tasks (Sheppard & Vernon, 2008), yet this has not been investigated or replicated with animal test batteries (see discussion by Flaim & Blaisdell, 2020). In contrast, the relationship between response inhibition and g has rarely been investigated in humans, but response inhibition tasks are almost always included in avian cognitive test batteries (Flaim & Blaisdell, 2020). Even when the cognitive domain does overlap, there are differences in the procedures used for humans versus nonhumans that can impede comparisons. Taking associative learning as an example, in humans an initial investigation using a simple associative learning task, where children had to learn which picture was associated with a reward, was not related to IQ scores (Plenderleith, 1956). More recent investigations have used the word-pairs task, where participants learn up to ten arbitrary pairs of words, like cat-pie, or the three-term contingency task where one word serves as a cue and the participant must learn three response words (Kaufman et al., 2009). These associative learning tasks show a positive relationship to g that scales with complexity, where the more complex three-term contingency task has a stronger relationship with g (Tamez et al., 2008; Williams & Pearlberg, 2006; but see Kaufman et al., 2009). In contrast, for mice and birds, a simple associative learning task, such as learning how to discriminate one cue from another to obtain a food reward, is related to the g like factor extracted in these species (Flaim & Blaisdell, 2020). The finding that associative learning is related to g across species, but different levels of difficulty are needed to reveal such a relationship, may be related to the experience of the subject. g is related to complexity, but it is also related to novelty, where novel tasks tend to have a high g loading (Carroll, 1993; Sternberg & Gastel, 1989). If animal subjects are naïve to any highly artificial experimental stimuli and procedures, the task may be sufficiently novel to explain why performance is related to g, despite the apparent simplicity. In contrast, when many humans are assessed, they have had years of experience in an educational setting with similar materials and task demands as the word-pairs and three-term contingency tasks. Therefore, for humans, task difficulty may be a more important factor for investigating associative learning and g. These results could indicate that task loading onto g is related to novelty, complexity, and associative learning across species, but further research is necessary to determine if there is a similar relationship between complexity and g in animals.

While animals may be relatively naïve to cognitive assessments compared to most human samples, there are other issues when comparing across species. In nonhuman research on g, the sample of animals tested is often homogeneous in some way (Shaw & Schmelz, 2017). In mice, only male subjects have been used and a a similar ‘home’ environment is used across studies (Kolata et al., 2008; but see Sauce et al., 2018), while for the wild bird subjects, like robins, collection is biased towards males and bold individuals (Shaw & Schmelz, 2017). If g is a robust phenomenon in animals, then it should replicate across members of the species that differ from the samples assessed thus far, for example in sex, personality, or environmental conditions. In addition, most experiments assess a small number of subjects. This can be overcome by using a consistent test battery, which makes it possible to pool results from multiple experiments, as demonstrated by Kolata et al. (2008). Utilizing a species that is more commonly investigated, either in the lab or across field sites, could also increase the number of subjects if multiple labs are willing to work together (Shaw & Schmelz, 2017). Thus, there could be improvements in both the test battery and sample characteristics, particularly for avian species. Tasks that assess clear cognitive domains, facilitate cross species comparisons, and have identified neural correlates should be favored. Species for which it is possible to obtain a large and diverse sample should also be favored, at least in these preliminary investigations of g in animals.

Given these arguments, it is surprising that pigeons have not been given a comprehensive test battery, given their long history as an animal model in psychology. Pigeons have excellent visual acuity and readily learn to peck visual stimuli in a touchscreen operant chamber, similar to procedures used to assess human and nonhuman primates (Wright et al., 2018; Zentall, 2021). Investigations of matching, interval timing, reaction time, memory, and many other cognitive domains show there are similarities in performance across pigeons and primates that indicate similar underlying mechanism at the psychological and neurobiological level (Colombo & Scarf, 2020; Güntürkün, 2005; Vickrey & Neuringer, 2000; Zentall, 2021). Methods for investigating memory, associative learning, and cognitive flexibility in particular have been well established, and the neural mechanisms supporting performance have been identified on some level. Similar to humans, performance on many cognitive tasks seems to depend on nidopallium caudolaterale (NCL) which is the avian equivalent to the mammalian prefrontal cortex (Güntürkün, 2005). For example, when assessing STM in the pigeon by requiring pigeons to remember a stimulus over a short delay to guide choice behavior, there is sustained neural activity in the NCL that relies on the neurotransmitter dopamine, similar to results found in nonhuman primates (Johnston et al., 2017; Karakuyu et al., 2007). Given this rich history, there are many tasks that could be included in a cognitive test battery for pigeons, but a few were selected as ideal for our purposes.

The tasks in the battery developed here were selected according to how well they assessed a specific cognitive domain, if the task facilitates cross-species comparisons, and if the neural substrates of performance had been identified (Diekamp et al., 2000; Flaim & Blaisdell, 2020; Izquierdo et al., 2017; Johnston et al., 2017; Karakuyu et al., 2007; Lissek et al., 2002; Vickrey & Neuringer, 2000). Ultimately, the pigeon cognitive test battery was designed to assess associative learning, cognitive flexibility, memory, and processing speed. Specifically, there were four tasks, symbolic match to sample (SMTS), serial reversal learning, delayed match to sample (DMTS), and a reaction time (RT) task. All the tasks were sufficiently sensitive to detect individual differences in performance, and all subjects completed at least two tasks in the battery (Table 1). Surprisingly, the correlation matrix from the test battery was not uniformly positive, and PCAs did not consistently yield a component similar to g. Potential procedural issues, the influence of age and experience, and the possibility that these results reflect a genuine difference between pigeons and other species are discussed.

Table 1. All subjects in the test battery and which tasks they completed, where 1 signifies they completed the task and 0 signifies they did not. Experience refers to the number of cognitive tasks completed before or between tasks in the cognitive test battery.

Test Battery			Tasks
Age	Experience	Name	Symbolic Match to Sample	Serial Reversal Learning	Delayed Match to Sample	Reaction Time	Total
17	11	Vonnegut	1	1	1	1	4
17	10	Dickinson	1	1	1	1	4
3	3	Bowser	1	1	1	1	4
3	2	Peach	1	1	1	1	4
3	2	Waluigi	1	1	1	1	4
3	1	Luigi	1	1	1	1	4
3	1	Mario	1	1	1	1	4
3	1	Shy Guy	1	1	1	1	4
3	1	Wario	1	1	1	1	4
17	6	Estelle	1	1	1	1	4
16	9	Jubilee	1	1	1	1	4
11	7	Herriot	1	1	1	1	4
17	9	Hawthorne	1	1	0	1	4
11	6	Goodall	1	1	1	1	4
0.5	0	Athena	1	1	1	1	4
0.5	0	Wenchang	1	1	1	1	4
16	9	Gambit	1	1	1	0	3
12	11	Darwin	0	1	1	1	3
12	6	Durrell	1	1	0	1	3
12	5	Cousteau	0	1	1	1	3
1	2	Itzamná	0	1	0	1	2
1	1	Odin	0	1	0	1	2
3	2	Yoshi	0	1	0	0	1
		Total	18	23	18	21

Subjects

Twenty-three pigeons served as subjects. The age at the start of the test battery ranged from 0.5–17 years old and ten were female. Subjects varied in how much experience they had with other cognitive tasks (Table 1). Subjects were individually housed in steel home cages with metal wire mesh floors in a vivarium. They were maintained at 80% of their free-feeding weight and provided free access to water and grit while in their home cages. Testing occurred at approximately the midpoint of the light portion of the 12-hour light-dark cycle. All procedures were approved by the UCLA Institutional Review Board.

Apparatus

Testing was conducted in a flat-black Plexiglas chamber (38 cm wide x 36 cm deep x 38 cm high). All stimuli were presented by computer on a color LCD monitor (NEC MultiSync LCD1550M) visible through a 23.2 x 30.5 cm viewing window in the middle of the front panel of the chamber. The bottom edge of the viewing window is 13 cm above the chamber floor. Pecks to the monitor were detected by an infrared touchscreen (Carroll Touch, Elotouch Systems, Fremont, CA) mounted on the front panel. A custom-built food hopper (Pololu, Robotics and Electronics, Las Vegas, NV) was located in the center of the front panel, its access hole flush with the floor. The food hopper contained a mixture of leach grain pigeon pellets and seed (Leach Grain and Milling). All experimental events were controlled and recorded with a variety of personal computers running the Windows 10 operating system. Stimuli were presented using the 2.7.11 version of Python with the psychopy toolbox, version 3.0.3 (Peirce, 2007).

Procedure

The tasks are described briefly here to emphasize the features that are relevant to the dependent measures ultimately included in the battery. Subjects did not receive the test battery in the same order, and it was not possible to fully counterbalance for order effects. Additionally, the time between tasks was not consistent across subjects. Subjects received one session per day, 3–7 days a week. All tasks were appetitive and used 3-s of access to a mixture of grain and seed as a reward.

Symbolic match to sample.

The SMTS was another measure of associative learning and was conceptually based on the arbitrary word-pairs task given to humans (Kaufman et al., 2009). The task is described in more detail in Flaim and Blaisdell (2023), but will be described briefly here. Instead of words, 18 subjects were presented with pictures of foods and animals which were associated through reinforcement. Subjects were trained on four pairs of pictures, where one of the pictures was always a food item while the other picture was always an animal, obtained from the food-pics database (Blechert et al., 2014). To associate pairs of pictures, trials had two phases, a sample phase and a choice phase. In the sample phase, one picture was shown in the center of the screen and is referred to as the sample. When subjects completed an observing response to the sample, pecking the picture ten times (FR10), the choice phase began. In the choice phase, the sample remained on the screen, and two comparison stimuli were presented below the sample on the left and right of the screen. If subjects pecked the correct comparison once (FR1), they received a food reward. If they pecked the incorrect comparison, a correction procedure was used where the trial repeated starting at the sample phase. Correction trials were not included in the analysis. The correct comparison was presented equally often on the left and right side of the screen and equally often with the three other incorrect comparison stimuli. Subjects were trained with this procedure until they were 80% accurate on each pair in a single session or until they had trained for 35 sessions. The number of sessions to reach criterion was used to measure associative learning ability (Table 1).

Serial reversal learning.

The serial reversal learning task was used to assess cognitive flexibility, being able to update behavior to reflect changes in environment (Izquierdo et al., 2017). Twenty-three subjects were trained with two stimuli, a blue or yellow circle, presented on the left and right side of the screen. One of the stimuli was always followed by a food reward (S+), while the other was not (S-). Subjects had to make three successive pecks (FR3) to indicate their choice and end the trial. When subjects were selecting the reinforced stimulus with 90% accuracy on two consecutive sessions, on the next session the contingency was reversed. The stimulus that was the S- was now the S + and vice versa. Subjects were trained on five reversals and performance on the first session of each reversal was used to measure cognitive flexibility (Table 1). Due to computer errors, performance on the first and second reversal was only measured for 21 subjects and performance on the fifth reversal was only measured for 22 subjects.

Delayed match to sample.

The DMTS task was used to assess memory, or more specifically the ability to maintain a memory of a stimulus over a short delay (Kangas et al., 2011). The task is described in more detail in Flaim and Blaisdell (2023), but will be described here briefly. Eighteen subjects were trained using a procedure similar to the SMTS task. There are three key differences between the DMTS and the SMTS task, the size of the stimulus set, what determined the correct comparison, and the delay period. Subjects were trained with two stimuli, a red circle and a green circle. First subjects were trained without a delay so the rule determining which comparison was correct could be learned. Similar to the SMTS, during this initial training each trial had two phases, a sample phase and a choice phase. During the sample phase, one stimulus was presented in the center location. After subjects completed the observing response (FR10) to the sample, the choice phase began and the two comparison stimuli were presented. One of the comparisons matched the sample, while the other did not. If subjects pecked (FR1) the matching comparison, they received a food reward. If they pecked the nonmatching comparison, a correction procedure was used where the trial repeated starting at the sample phase. The correction trials were not included in the analysis. Each stimulus served as the sample an equal number of times and the correct comparison was presented equally often on the left and right side of the screen. When subjects reached criterion on the initial training, 80% correct on two consecutive sessions, they stated the DMTS.

In the DMTS, each trial had three phases, sample, delay, and choice. The sample phase is the same as the training described above. Once subjects had completed the observing response, the sample was removed from the screen, and the delay phase began. The delay could be 0, 2, 4, or 8 s, and each delay length was presented an equal number of times with each combination of correct comparison stimulus and location After the delay had elapsed, only the comparison stimuli were presented. Again, if subjects pecked the matching comparison, they received a food reward. If subjects pecked the nonmatching comparison, the trial ended without reinforcement. The correction procedure was not used during the DMTS. Subjects were trained on this procedure for 30 sessions. Accuracy over the last three sessions for all delay lengths was used to assess performance.

Reaction time task.

This task was used to assess response speed with a procedure that relied on detecting a change in the stimulus display (Sheppard & Vernon, 2008). Twenty-two subjects were trained and took part in four different experiments with slight procedural differences overall that are described in more detail in Flaim et al. (2023; n = 4–7 for each experiment), but the procedure described here was the same across experiments. The stimulus display consisted of nine circular stimuli that could either be a white outline or completely filled with white. One of the stimuli was in the center of the screen, near the bottom of the viewing window. This stimulus was the home key. The remaining eight stimuli were arranged in a semi-circle around the home key and were potential targets (PTs). Trials had two phases, a home key phase and a choice phase. During the home key phase, the home key was filled with white and subjects had to peck the home key three times on average (VR3 +/- 2). When subjects completed the peck requirement to the home key, the choice phase began. During the choice phase, the home key became a white outline and one of the PTs was filled with white and became the target. If subjects pecked (FR1) the target, they received a food reward. If subjects pecked a PT it was counted as an error of commission, but if subjects did not peck the target or a PT within 5 s, it was counted as an error of omission. Only trials where subjects successfully pecked the target were included in analysis (error rates were very low). The target appeared in each location an equal number of times. Each session either had 20 (n = 10) or 24 trials (n = 12) with eight PTs, and subjects were trained for 10 and 9 sessions, respectively. Median reaction time (RT) collapsed across all sessions was used as the dependent measure. The number of trials included in the analysis for each subject ranged from 184–216 depending on how many trials had to be excluded due to errors.

Data Analysis

Comparing performance across the tasks was difficult since the different dependent measures do not use the same direction to indicate better or worse performance or the same scale. For the serial reversal learning and DMTS procedures, better performance was indicated with a larger number and worse performance is indicated with a smaller number. For the SMTS and RT tasks, performance was coded in the opposite direction, where a larger number indicates worse performance. For these two tasks, the data were reverse coded, so all results were comparable where a larger number indicates better performance. Additionally, the scale of the dependent measure differs across tasks. For example, the SMTS task ranges from 8–35, while the serial reversal learning and DMTS tasks were bound between 0–1. Since the range of the data can impact correlations, the data were normalized (observed value – mean value / the standard deviation) so the range was comparable across tasks.

For the DMTS, where performance was collapsed across three sessions, the test-retest reliability of the performance was assessed with a Pearson correlation. Similarly, a Pearson correlation was also used to assess performance over different conditions within a task. This analysis was used to compare performance for the different delay lengths in the DMTS task and the different reversals in the serial reversal learning task. To assess reliability for the RT time task, where performance was collapsed across 9 or 10 sessions, a Pearson correlation was used to compare performance at the beginning (first three sessions) and end (last three sessions) of training. For the RT task, because subjects experienced different procedures, an ANOVA was also used to determine if there were any significant differences in median RT across the experiments.

A Pearson correlation with a Bonferroni correction was used to compare performance across tasks. Age and experience were included as variables that were related to performance (Table 1). PCAs with an unrotated factor solution were used to determine if variance in performance in the test battery could be accounted for with a single component. All statistical analyses were performed using JASP version 0.16.1.

Individual Cognitive Tasks

Symbolic Match to Sample

The number of sessions needed to reach criterion, 80% accuracy on all four pairs in a single session, was used as the dependent measure. If subjects did not reach criterion, the maximum number of sessions, 35, was used. The number of sessions required ranged from 8–35 (M = 20.5, SD = 9.13). The data were reverse coded, so a larger number indicated better performance, and then normalized.

Serial Reversal Learning

Performance on the first session of the initial discrimination and each reversal was used as the dependent measure. Performance on the initial discrimination and the first and second reversals had weak and nonsignificant correlations with all measures. Performance on the third and fifth and fourth and fifth reversals were significantly, positively correlated with each other. While performance on the third and fourth reversal was also strongly, positively correlated, this did not survive a Bonferroni correction (Table 2). This indicated that the initial discrimination and the first and second reversals were not assessing the same ability or cognitive domain as the third, fourth, and fifth reversals, which is also supported by previous research (Izquierdo et al., 2017). Due to the strong correlations between performance on the fourth and fifth reversals, an aggregate measure of performance was created by collapsing across the two conditions. While performance on the third and fifth reversals also had a strong correlation, this may have been due to both conditions having the same reinforced stimulus.

Table 2

Correlation matrix between the measures of the serial reversal learning task. The number inside the parenthesis is the sample size. Italicized values indicate the result was significant before correcting for multiple comparisons, while bolded values indicate the result was significant after a Bonferroni correction.
Reversal Learning
		Initial	First	Second	Third	Fourth
First	r	0.013 (20)	--
	p	0.957
Second	r	0.031 (22)	0.189 (21)	--
	p	0.891	0.413
Third	r	-0.05 (22)	0.04 (21)	0.154 (23)	--
	p	0.825	0.863	0.483
Fourth	r	0.308 (22)	0.083 (21)	0.001 (23)	.572 (23)	--
	p	0.162	0.72	0.995	0.004
Fifth	r	0.108 (21)	0.001 (20)	0.101 (22)	.712 (22)	.623 (22)
	p	0.641	0.996	0.656	> 0.001	0.002

Delayed Match to Sample

Performance collapsed over the last three sessions (28, 29, and 30) of training was used as the dependent measure. For one subject, Goodall, the 27, 28, and 29th sessions were used due to a computer error in the 30th session. Performance with the 2-second delay was the most reliable with an average correlation of .809 across the last three sessions. Performance with the 4-second delay was also reliable, with an average correlation of .713. Performance at the 0- and 8-second delays were not as reliable, with an average correlation of .442 and .554 respectively. Performance on the 2-second and 4-second delay were also significantly, positively correlated with each other (r = .783, n = 18, p < .001), thus an aggregate measure of performance was collected by collapsing across the two conditions.

Reaction Time

While explained in more detail in Flaim et al. (2023), justification for the dependent variable used will briefly be given here. Only trials where subjects correctly pecked the target were used in the analysis, because errors were infrequent and difficult to interpret. To investigate potential effects of task experience, the mean median RT from the first and last 3 sessions of training were compared. There was no evidence of any consistent or reliable changes with task experience across any of the experiments. A Pearson correlation between the mean median RT of the first and last 3 sessions was significant (r = .921, n = 21, p < .001), indicating that performance was reliable across sessions. Therefore, the median RTs collapsed across all sessions of training were used in the present analyses. A one-way ANOVA confirmed there were no statistically significant differences in mean median RT based on which experiment a subject experienced (F < 1). Finally, it should be noted one subject, Yoshi, was excluded from the analyses since the number of errors classified it as an outlier (Flaim et al., 2023).

Cognitive Test Battery

Correlation matrix

A Pearson correlation was used to determine if there was any relationship in performance across the different cognitive tasks. Before the analysis, the data from the SMTS and RT tasks were reversed coded so for all tasks a larger number indicated better performance. The data were also normalized so there would be a similar range in scale across all tasks. Tasks with multiple dependent measures were analyzed in two ways, with the aggregate measure where performance was collapsed across highly correlated conditions as described above, and with the individual task performance that composed the aggregate measure. For the DMTS, performance on the 2- and 4-second delay were included, and for the reversal learning task, performance on the fourth and fifth reversal were included in the analyses. Finally, age and experience were also included in the analysis since they could have a potential relationship with performance.

The correlation matrix with the aggregate measures was not uniformly positive across the cognitive tasks (Table 3). Performance on the SMTS and RT were positively correlated, but had weak to negative correlations with the other tasks. Similarly, performance on the DMTS and reversal learning tasks were positively correlated, but had weak correlations with the other tasks. None of these correlations survived a Bonferroni correction (α = .008). The average correlation between the cognitive tasks was .226.

Age and experience had a strong, significant positive correlation with each other and thus had a similar relationship with the cognitive tasks. Age and experience were negatively correlated with almost all cognitive tasks, though only the correlation between age and the SMTS task was significant before correction. Only the DMTS had no relationship with age or experience (Flaim & Blaisdell, 2023). A partial correlation controlling for age had little effect on the correlations between the cognitive tasks (Table 3).

Table 3. Correlation matrix between the measures of the cognitive test battery, age, and experience. An aggregate measure was used for the reversal learning and delayed match to sample (DMTS) tasks. The values in the lower half of the triangle are zero-order correlations and the number inside the parenthesis is the sample size. The values in the upper half of the triangle are partial correlations controlling for age and the number inside the parenthesis are the degrees of freedom. Bolded values indicate the result was significant before correcting for multiple comparisons and italicized values indicate results that were no longer significant after controlling for age.

Pearson Correlation
Variable		SMTS	Reversal Learning	DMTS	RT	Experience
SMTS	r		0.117 (15)	0.072 (13)	0.345 (14)	0.299 (15)
	p		0.654	0.799	0.19	0.243
Reversal Learning	r	0.201 (18)		0.496 (14)	.087 (17)	0.308 (19)
	p	0.423		0.051	0.724	0.175
DMTS	r	0.074 (16)	0.495 (17)		-.019 (14)	0.113 (15)
	p	0.785	0.043		0.481	0.665
RT	r	0.524 (17)	0.227 (20)	-0.166 (17)		0.245 (18)
	p	0.031	0.336	0.525		0.298
Experience	r	-0.404 (18)	-0.089 (22)	0.052 (18)	-0.234 (21)
	p	0.096	0.694	0.836	0.308
Age	r	-0.491 (18)	-0.218 (22)	0.011 (18)	-0.35 (21)	0.929 (23)
	p	0.038	0.33	0.964	0.119	< .001

Similar results were obtained when using the individual measures of performance for the reversal learning and DTMS tasks (Table 4). Performance on the fifth reversal had a stronger correlation with the matrix displays task and 2- and 4-second delay from the DMTS task compared to the fourth reversal. Performance on the 4-second delay had more positive correlations with the other cognitive measures compared to the 2-second delay. The average correlation between tasks across the matrix was .258. None of the correlations across tasks were significant after a Bonferroni correction (α = .002). Similar to the analysis with the aggregate measures, age and experience were negatively correlated with all cognitive measures, except for the 4-second delay condition. Controlling for age did not substantially alter the correlations between the cognitive tasks (Table 4).

Table 4

Correlation matrix between the measures of the cognitive test battery, age, and experience. The values in the lower half of the triangle are zero-order correlations and the number inside the parenthesis is the sample size. The values in the upper half of the triangle are partial correlations controlling for age and the number inside the parenthesis are the degrees of freedom. Bolded values indicate the result was significant before correcting for multiple comparisons and italicized values indicate results that were no longer significant after controlling for age. Bolded and underlined values indicate the measures were from the same task.
Variables		SMTS	Fourth Reversal	Fifth Reversal	2 Sec Delay	4 Sec Delay	RT	Experience
SMTS	r		0.093 (15)	0.123 (15)	-0.021 (13)	0.16 (13)	0.345 (14)	0.299 (15)
	p		0.722	0.637	0.941	0.569	0.19	0.243
Fourth Reversal	r	0.062 (18)		0.65 (19)	0.306 (15)	0.45 (15)	-0.006 (15)	0.314 (20)
	p	0.806		< .001	0.233	0.07	0.979	0.155
Fifth Reversal	r	0.307 (18)	0.623 (22)		0.422 (14)	0.538 (14)	0.144 (17)	0.248 (19)
	p	0.216	0.002		0.104	0.032	0.557	0.278
2 Sec Delay	r	0.026 (16)	0.294 (18)	0.421 (17)		0.792 (15)	-0.095 (14)	0.018 (15)
	p	0.925	0.237	0.092		< .001	0.727	0.946
4 Sec Delay	r	0.116 (16)	0.456 (18)	0.485 (17)	0.783 (18)		-0.265 (14)	0.194 (15)
	p	0.669	0.057	0.048	< .001		0.322	0.455
RT	r	0.524 (17)	0.06 (21)	0.323 (20)	-0.062 (17)	-0.252 (17)		0.245 (18)
	p	0.031	0.796	0.165	0.814	0.329		0.298
Experience	r	-0.417 (18)	0.061 (23)	-0.224 (22)	-0.034 (18)	0.161 (18)	-0.258 (22)
	p	0.085	0.781	0.317	0.895	0.523	0.259
Age	r	-0.491 (18)	-0.062 (23)	-0.351 (22)	-0.061 (18)	0.08 (18)	-0.35 (21)	0.929 (23)
	p	0.038	0.78	0.11	0.811	0.752	0.119	< .001

Principal component analysis

A total of ten PCAs were conducted. To investigate the effect of missing data, PCAs were performed using pairwise and listwise deletion (Dray & Josse, 2015). For listwise deletion, this meant that only the subjects that had completed all four tasks (n = 15, Table 1), were included in the analyses. Since some of the tasks had multiple dependent measures or an aggregate measure, five PCAs were conducted to ensure that all tasks were equally represented and to determine if the component structure was robust. Finally, because the data were normalized, the PCA was based on the covariance matrix (Jolliffe & Cadima, 2016) and was unrotated so the first component could account for the maximum amount of variance. By convention, only Eigenvalues larger than 1 were retained.

Table 5

The loadings and percentage of variance explained for each principal component, with two different methods of handling missing data and different dependent measures. The top row of analyses used an aggregate measure from the delay match to sample (DMTS) and reversal learning task, while the subsequent analyses use the individual measures from the tasks. For the symbolic match to sample (SMTS) and RT (reaction time) tasks, the same measures are used for all analyses.
Principal Component Analysis
	Pairwise deletion		Listwise deletion (n = 16)
Task	PC1	PC2	PC1	PC2
SMTS	0.75		0.88
RT	0.68	-0.59	0.86
Reversal Learning	0.72	0.49		0.78
DMTS	0.42	0.81		0.87
Eigenvalue	1.71	1.39	1.65	1.48
% Variance Explained	42.8	34.6	41.3	36.8
SMTS	0.87		0.89
RT	0.86		0.9
Fourth Reversal		0.81		0.8
2-Sec Delay		0.77		0.8
Eigenvalue	1.54	1.29	1.62	1.3
% Variance Explained	38.5	32.2	40.6	32.6
SMTS	0.8		0.78	0.48
RT	0.9		0.9
Fourth Reversal		0.83		0.75
4-Sec Delay		0.84	-0.42	0.74
Eigenvalue	1.54	1.48	1.67	1.39
% Variance Explained	38.6	37.2	41.7	34.6
SMTS	0.76		0.82
RT	0.75	-0.45	0.82
Fifth Reversal	0.75	0.43	0.6	0.6
2-Sec Delay		0.85		0.836
Eigenvalue	1.83	1.23	1.77	1.31
% Variance Explained	45.7	30.8	44.2	32.7
SMTS	0.78		0.85
RT	0.7	-0.61	0.84	-0.41
Fifth Reversal	0.78	0.41	0.55	0.64
4-Sec Delay		0.88		0.9
Eigenvalue	1.82	1.37	1.75	1.4
% Variance Explained	45.5	34.2	43.7	35.1

For all the PCAs, two components with Eigenvalues larger than 1 were extracted (Table 5). For the analyses using pairwise deletion, the task loadings depended on which measure of reversal learning was included in the analysis. When the aggregate measure was included, all tasks positively loaded onto the first component, which could account for 42.8% of the variance in performance. While the aggregate measure of reversal learning, fifth reversal, and DMTS positively loaded onto the second component, the RT task negatively loaded onto this component, which could account for 30.8–37.2% of the variance in performance. When the fourth reversal was included in the analyses, the fourth reversal and DMTS only positively loaded onto the second component. When the fifth reversal was included, the fifth reversal positively loaded onto both components, while the DMTS only positively loaded onto the second component.

For the analyses using listwise deletion, the five PCAs showed a similar pattern, where performance on the SMTS and RT tasks positively loaded onto the first component, which could account for 40.6–44.2% of the variance, while performance on the reversal learning and DMTS tasks positively loaded onto the second component, which could account for 32.7–36.8% of the variance. The SMTS task positively loaded onto the second component when the fourth reversal and 4-sec delay were included in the PCA. RT never loaded positively onto the second component and had a negative loading when the fifth reversal and 4-sec delay were included in the PCA. The DMTS task never positively loaded onto the first component and even had a negative loading on the first component when the analyses included the 4-sec delay and fourth reversal. For reversal learning, the aggregate measure and fourth reversal never loaded onto the first component, while the fifth reversal always had a positive loading on both components (Table 5).

This is the first time that cognitive performance in the pigeon has been systematically investigated using a test battery. The battery was created with the intention of assessing different cognitive domains, including associative learning, cognitive flexibility, STM, and RT. We predicted that performance would positively correlate across tasks. In contrast to our predictions, the correlation matrix was not uniformly positive and there were indications that tasks were forming two clusters. The reversal learning and DMTS tasks showed stronger positive correlations, while performance on the SMTS and RT tasks were only positively correlated with each other. Even though age and experience were negatively related to almost all cognitive tasks, controlling for age did not have a substantial impact on the overall pattern of correlations between the cognitive tasks.

The PCAs confirmed the clustering seen in the correlation matrix. When pairwise deletion was used to handle missing data, all tasks had a positive loading on the same factor in only one out of the five PCAs. When listwise deletion was used, there was no evidence for g, instead the cognitive tasks differentially loaded onto the two components. The SMTS and RT tasks always had a positive loading onto the first component and no, or a negative loading on the second component, with one exception. The reversal learning and DMTS measures had variable loadings on the first component, but always positively loaded onto the second component. A negative loading indicates that these tasks capture the opposite of what the second component represents (Bro & Smilde, 2014). For the DMTS, performance always had strong positive loading onto the second component. When the DMTS had a positive loading on the first component, it was always weaker than the loading on the second component, and it was more common for the DMTS to not load onto the first component or have a negative loading. For the serial reversal learning task, the loading was more variable depending on if performance on the fourth or fifth reversal was included in the analyses. The fourth reversal never loaded onto the first component. The fifth reversal always had a positive loading onto the first component, but which component had a stronger loading depended on how the missing data were handled. For the pairwise deletion, there was a stronger loading on the first component, but with the listwise deletion, there was either an equal loading with or a stronger loading on the second component. Listwise deletion intensifies the divide between the SMTS and RT tasks primarily reflecting one component while the reversal learning and DMTS tasks reflected another. This divide is present to a smaller degree when pairwise deletion is used. While listwise deletion is viewed as less reliable since the analyses are conducted with a smaller sample size, for these data set there is not an extreme difference in sample size for listwise (n = 16) compared to pairwise (n = 16–21) deletion (Van Ginkel et al., 2014). In addition, the correlation matrix supports separate clusters of cognitive tasks. Therefore, it is worthwhile to explore what these two components represent.

Why do the tasks in the battery show these patterns of results? It could be due to an overall difference in strategy that could be used for the two sets of tasks. In the serial reversal learning and DMTS tasks, it was possible to use a more general rule-based strategy. For serial reversal learning subjects could use a ‘win-stay, lose-shift’ rule, and for DMTS subjects could, as the name implies, use a matching rule. For the SMTS and RT tasks, it was not possible to use a rule-based strategy. In the SMTS task, the pairs were selected because they did not have any consistent perceptual feature that could be used to guide choice behavior. In the RT task, the location of the target was pseudorandomized and it would have been difficult for subjects to predict where it would appear. Therefore, one interpretation of the components would be tasks that rely on memorization or associative learning (component 1) and tasks where it is possible to use a rule-based strategy (component 2). Yet, this is not wholly satisfactory given the difference in loadings for the fourth and fifth reversals. While the fourth and fifth reversal positively loaded onto the second ‘rule-based’ component, the fifth reversal always loaded positively on the first ‘memorization/associative learning’ component as well. In contrast, the fourth reversal never positively loaded onto this first component. According to the proposed interpretation, this would indicate that by the fifth reversal, a general rule has less control over performance. Instead of the second component reflecting a rule-based strategy, it could reflect inhibitory control. Arguably, subjects would perform better on the choice RT task with poor inhibitory control since they could react faster to a change in their environment. Yet poor inhibitory control would likely lead to worse performance on the serial reversal learning task since subjects would need to inhibit their peck response to the previously reinforced stimulus when presented with a new contingency.

It is also possible these two components reflect differences in automaticity. Human research has indicated that skill or task learning occurs in three phases, where the initial learning phase requires effortful cognitive processes while the final phase relies on automated responses (Ackerman, 1988). Progression though these phases partially depends on the complexity and consistency of the task (Ackerman, 1988). Component 1 could reflect tasks that have become automatic while component 2 reflects tasks that are not yet automatic. It is consistent with the human literature that the choice RT task would quickly become automated due to its high level of stimulus-response compatibility and consistent task demands. This could also account for the difference between the fourth and fifth reversal in the serial reversal learning task. As training continues, performance is more likely to rely on automatic responses (Ackerman, 1988).

Finally, these components could also reflect differences in which tasks are sensitive to age-related declines. Component 1 reflects tasks that show age related declines in performance, while component 2 reflects tasks where performance does not change with age. Performance on the SMTS (associative) and choice RT tasks was negatively correlated with age, which is consistent with the human literature. The potential reasons for age related declines in the SMTS task specifically are described in more detail in Flaim and Blaisdell (2023). Research with humans has demonstrated that associative learning, as measured with the arbitrary word pairs task, is related to processing speed and that performance declines with age (Rast & Zimprich, 2009; Salthouse, 1994). This indicates that pigeons may have similar age-related changes in cognition compared to humans, but additional research is needed. Even though these interpretations are presented individually, they are not mutually exclusive. It is likely that more specialized cognitive abilities, like inhibitory control, influence rate of automation or are also influenced by age-related changes in performance.

While these interpretations are still speculative, the most striking result from the test battery is that there is more evidence for a two-component structure rather than a g-like factor. The potential two-component structure contradicts our predictions and many of the previous results with other species (Flaim & Blaisdell, 2020). While the results from this test battery are not a definitive conclusion on the structure of pigeon cognition, especially given the relatively small sample size, it would be helpful to identify which features of the sample or test battery led to these results. There are three primary issues that will be discussed. The first and most obvious issue is that the statistical analyses are underpowered (Bonett & Wright, 2000; Mundfrom et al., 2005). Nevertheless, a g like factor has been found in samples of similar size (Isden et al., 2013; Shaw et al., 2015). Thus, while this is an issue across a few animal cognitive test batteries, it doesn’t necessarily preclude finding the results we predicted.

The second issue is that the tasks in this cognitive battery required more sessions of training compared to the cognitive test batteries given to other species, including humans (Johnson et al., 2004). For example, it took robins an average of 17 days to complete a five-task battery (Shaw et al., 2015). In contrast, subjects needed an average of 84.45 days (SD = 12.78) to complete the serial reversal learning, SMTS, DMTS, and RT tasks, not including preliminary training. It could be that the amount of training and time each subject spent on the task explains this factor structure, rather than reflecting a genuine difference between pigeons and other species. Yet it is not clear how differences in training could cause this specific cognitive structure or prevent a g like factor from being found, particularly when the tasks were specifically investigated for sensitivity to individual differences. Even with the length of training, performance on tasks in the battery did not show ceiling or floor effects (Schubiger et al., 2020; Völter et al., 2018). When possible, the data were also analyzed for reliability and the results indicated that the measures used in analyses were moderately stable over a short time frame. While noncognitive traits like persistence were not assessed, many of these tasks were selected because previous research indicated that they were cognitively demanding (Izquierdo et al., 2017; Kaufman et al., 2009; Zentall & Smith, 2016). The least cognitively demanding task, RT, was selected because of the consistent relationship between g and RT in humans (Sheppard & Vernon, 2008). Therefore, it seems unlikely that either the length of training or the tasks themselves could prevent a g like factor from being consistently extracted from a PCA.

The third issue is the difference in experience with cognitive tasks that our pigeons have compared to the other species primarily discussed thus far. The mice and avian cognitive test batteries have used experimentally-naïve subjects, which would make the testing apparatus and subsequent experimental conditions very novel. The subjects in this test battery, like many other pigeon and nonhuman primate research laboratories, had different amounts of experience with other cognitive tasks in the touchscreen operant chamber. This meant that all subjects had been exposed to different stimuli, peck requirements, and reinforcement rates, which reduces the novelty of the tests in the battery, as well as reducing the novelty of the test context, handling cues, home environment, etc. (Table 1). As mentioned in the introduction, novelty is a factor for why some tasks show a strong g loading, but this is not true for all tasks used in human research. For example, vocabulary tests also have a strong g loading and primarily rely on retrieving previously learned knowledge under familiar conditions (Colom et al., 2002; Gignac, 2015). In addition, human FSIQ are typically administered using formats commonly found in Western educational settings (Clark et al., 2016), such that the apparatus and general procedure are familiar to most of the participants, similar to the pigeon subjects in this experiment. So, if the prior experience with the apparatus and general experimental procedure prevented a g like factor in pigeons, it would have major implications when comparing g across humans as well. Ultimately, however, the difference between naïve and experienced subjects is an unlikely explanation for these differences since experience with other cognitive tasks had a relatively weak relationship with performance on the cognitive test battery (Table 2, 3).

Differences in the test battery and experimental history of the subjects are not satisfactory explanations for why we did not extract a robust g like factor. Is there a satisfactory explanation at the species level? While previous research has demonstrated striking similarities in performance between pigeons and primates, the parameters for each species were different (Colombo & Scarf, 2020). Pigeons need more stimuli or examples to show evidence for general rule learning compared to primates. For example, Wright and Katz (2006) have investigated same/different concept learning in Rhesus and capuchin monkeys and pigeons, using similar procedures and identical stimuli. Their criterion for full abstract concept learning is that performance with completely novel stimuli must match performance with previously trained stimuli. Initially, all subjects were trained with a set of eight pictures until they were 80% accurate, then they were tested with novel pictures. The set size of the pictures doubled if subjects failed to show full concept learning at test. Rhesus and capuchin monkeys had full concept learning with a set size of 128 pictures, whereas most pigeons needed a set size 256 pictures, and one pigeon needed 1024! It should be noted that none of the species here showed any evidence for concept learning with the initial set of 8 pictures. So, it is possible that all subjects initially learned the task by memorizing each initial item pair then all gradually transferred to using an abstract matching concept, but pigeons were the slowest to transfer. The obvious question is why are pigeons slower to transfer. One answer is because pigeons are birds, but the type of bird may be an important qualifier. Subsequent investigations have shown that nutcrackers and black-billed magpies, members of the corvid family, show full concept learning with 128 pictures, the same as the monkeys (Wright et al., 2018). It has been stated that pigeon’s ‘preferred strategy’ is the concrete one, where subjects will use each unique stimulus configuration to guide behavior as opposed to a more flexible, stimulus independent rule (Wright, 1997, p. 119). This seems to apply to a wide variety of tasks, but the goal of these experiments is to figure out how to break the use of a concrete strategy to allow for the use of a relational or conceptual strategy (Colombo & Scarf, 2020; Wright, 1997; Wright & Katz, 2006). It would be helpful to understand why pigeons tend to use a concrete strategy over an abstract one (O'Donoghue et al., 2022), as it may be related to the two-component structure extracted in this experiment.

Further speculations on why pigeons have a different cognitive structure than other species should be restrained until these results are replicated. While there are clear strengths to the tasks used in this battery, there are also weaknesses. Future, stronger test batteries should assess a broader array of cognitive domains. Spatial reasoning in particular would be an excellent addition since there have been investigations directly comparing the abilities of pigeons and humans in a variety of paradigms (Hollard & Delius, 1982; Spetch et al., 1996). Non-cognitive factors, like persistence and neophobia, should also be assessed. While it seems unlikely to explain the results obtained here, non-cognitive factors have been shown to differentially impact performance on seemingly cognitive tasks and should be explicitly accounted for (Carere & Locurto, 2011; Cole & Quinn, 2012; Delacoux & Guenther, 2023; Isden et al., 2013; Shaw & Schmelz, 2017). Finally, and on a more practical note, ideally the entire test battery should take far less time to complete. The amount of time to train and test subjects limits the number of tasks that can be included and how feasible it is for other labs to replicate the results. Investigating g across a variety of species could help determine if there are consistent neuroanatomical features present in species that exhibit a g factor compared to species that do not. Ultimately this test battery is an interesting step towards understanding the general cognitive abilities of the pigeon. Future investigations are sure to yield insights about the structure of general cognitive abilities.

Ethical Approval

All procedures were approved by the UCLA Institutional Review Board.

Competing Interests

The authors have no relevant financial or non-financial interests to disclose.

Author Contributions

M.F. and A.P. developed the methodology for the experiment. M.F. conceptualized the idea, wrote the original draft, conducted the investigation and formal analysis, and oversaw project administration. A.P. provided resources, funding, supervision, and reviewed and edited the manuscript.

Funding

Support for this research was provided by National Science Foundation grant BCS-1844144 (Aaron P. Blaisdell).

Data Availability

Data is available upon request.

Ackerman, P. L. (1988). Determinants of individual differences during skill acquisition: Cognitive abilities and information processing. Journal of Experimental Psychology: General, 117(3), 288. http://dx.doi.org/10.1037/0096-3445.117.3.288
Anderson, R. C., Searcy, W. A., Peters, S., Hughes, M., DuBois, A. L., & Nowicki, S. (2017). Song learning and cognitive ability are not consistently related in a songbird. Animal Cognition, 20(2), 309–320. http://dx.doi.org/10.1007/s10071-016-1053-7
Ashton, B. J., Ridley, A. R., Edwards, E. K., & Thornton, A. (2018). Cognitive performance is linked to group size and affects fitness in Australian magpies. Nature, 554(7692), 364–367. http://dx.doi.org/10.1038/nature25503
Blechert, J., Meule, A., Busch, N. A., & Ohla, K. (2014). Food-pics: an image database for experimental research on eating and appetite. Frontiers in Psychology, 5, 617. http://dx.doi.org/10.3389/fpsyg.2014.00617
Blum, D., & Holling, H. (2017). Spearman's law of diminishing returns. A meta-analysis. Intelligence, 65, 60–66. http://dx.doi.org/10.1016/j.intell.2017.07.004
Bonett, D. G., & Wright, T. A. (2000). Sample size requirements for estimating Pearson, Kendall and Spearman correlations. Psychometrika, 65(1), 23–28. http://dx.doi.org/10.1007/BF02294183
Boogert, N. J., Anderson, R. C., Peters, S., Searcy, W. A., & Nowicki, S. (2011). Song repertoire size in male song sparrows correlates with detour reaching, but not with other cognitive measures. Animal Behaviour, 81(6), 1209–1216. http://dx.doi.org/10.1016/j.anbehav.2011.03.004
Bray, M. A., Kehle, T. J., & Hintze, J. M. (1998). Profile analysis with the Wechsler Scales: Why does it persist?. School Psychology International, 19(3), 209–220. http://dx.doi.org/10.1177/0143034398193002
Bro, R., & Smilde, A. K. (2014). Principal component analysis. Analytical Methods, 6(9), 2812–2831. http://dx.doi.org/10.1039/C3AY41907J
Burkart, J. M., Schubiger, M. N., & van Schaik, C. P. (2017). The evolution of general intelligence. Behavioral and Brain Sciences, 40. http://dx.doi.org/10.1017/S0140525X16000959
Carere, C., & Locurto, C. (2011). Interaction between animal personality and animal cognition. Current Zoology, 57(4), 491–498. http://dx.doi.org/10.1093/czoolo/57.4.491
Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-analytic studies. Cambridge University Press.
Chuderski, A. (2013). When are fluid intelligence and working memory isomorphic and when are they not?. Intelligence, 41(4), 244–262. http://dx.doi.org/10.1016/j.intell.2013.04.003
Clark, C. M., Lawlor-Savage, L., & Goghari, V. M. (2016). The Flynn effect: A quantitative commentary on modernity and human intelligence. Measurement: Interdisciplinary Research and Perspectives, 14(2), 39–53. http://dx.doi.org/10.1080/15366367.2016.1156910
Cole, E. F., & Quinn, J. L. (2012). Personality and problem-solving performance explain competitive ability in the wild. Proceedings of the Royal Society B: Biological Sciences, 279(1731), 1168–1175. http://dx.doi.org/10.1098/rspb.2011.1539
Colom, R., Abad, F. J., Garcıa, L. F., & Juan-Espinosa, M. (2002). Education, Wechsler's full scale IQ, and g. Intelligence, 30(5), 449–462. http://dx.doi.org/10.1016/S0160-2896(02)00122-8
Colom, R., Burgaleta, M., Román, F. J., Karama, S., Álvarez-Linera, J., Abad, F. J., Martínez, K., Quiroga, M. A., & Haier, R. J. (2013). Neuroanatomic overlap between intelligence and cognitive factors: Morphometry methods provide support for the key role of the frontal lobes. Neuroimage, 72, 143–152. http://dx.doi.org/10.1016/j.neuroimage.2013.01.032
Colombo, M., & Scarf, D. (2020). Are There Differences in" Intelligence" Between Nonhuman Species? The Role of Contextual Variables. Frontiers in Psychology, 11, 2072–2072. http://dx.doi.org/10.3389/fpsyg.2020.02072
Conway, A. R., & Kovacs, K. (2015). New and emerging models of human intelligence. Wiley Interdisciplinary Reviews: Cognitive Science, 6(5), 419–426
Conway, A. R., Cowan, N., Bunting, M. F., Therriault, D. J., & Minkoff, S. R. (2002). A latent variable analysis of working memory capacity, short-term memory capacity, processing speed, and general fluid intelligence. Intelligence, 30(2), 163–183. http://dx.doi.org/10.1002/wcs.1356
de Winter*, J. C., Dodou*, D., & Wieringa, P. A. (2009). Exploratory factor analysis with small sample sizes. Multivariate Behavioral Research, 44(2), 147–181. http://dx.doi.org/10.1080/00273170902794206
Deary IJ. Looking down on human intelligence: From psychometrics to the brain. Oxford, England: Oxford University Press; 2000.
Deary, I. J., Penke, L., & Johnson, W. (2010). The neuroscience of human intelligence differences. Nature Reviews Neuroscience, 11(3), 201–211. http://dx.doi.org/10.1038/nrn2793
Delacoux, M., & Guenther, A. (2023). Stressfulness of the design influences consistency of cognitive measures and their correlation with animal personality traits in wild mice (Mus musculus). Animal Cognition, 26(3), 997–1009. http://dx.doi.org/10.1007/s10071-023-01748-3
Detterman, D. K. (1991). Reply to Deary and Pagliari: Is g intelligence or stupidity?. Intelligence, 15(2), 251–255. http://dx.doi.org/10.1016/0160-2896(91)90034-B
Diekamp, B., Kalt, T., Ruhm, A., Koch, M., & Güntürkün, O. (2000). Impairment in a discrimination reversal task after D1 receptor blockade in the pigeon" prefrontal cortex". Behavioral Neuroscience, 114(6), 1145. http://dx.doi.org/10.1037/0735-7044.114.6.1145
Divac, I., Mogensen, J., & Björklund, A. (1985). The prefrontal ‘cortex’ in the pigeon. Biochemical evidence. Brain Research, 332(2), 365–368. http://dx.doi.org/10.1016/0006-8993(85)90606-7
Dray, S., & Josse, J. (2015). Principal component analysis with missing values: a comparative survey of methods. Plant Ecology, 216(5), 657–667. http://dx.doi.org/10.1007/s11258-014-0406-z
Flaim, M., & Blaisdell, A. P. (2020). The comparative analysis of intelligence. Psychological Bulletin, 146(12), 1174. http://dx.doi.org/10.1037/bul0000306
Flaim, M., & Blaisdell, A. P. (2023). The effect of age on delay performance and associative learning tasks in pigeons. Learning & Behavior, 1–14. http://dx.doi.org/10.3758/s13420-022-00565-x
Flaim, M., Guo, J., & Blaisdell, A. P. (2023). Choice reaction time in pigeons fails to increase as predicted by Hick’s law. Behavioural Processes, 206, 104838. http://dx.doi.org/10.1016/j.beproc.2023.104838
Fraenz, C., Schlüter, C., Friedrich, P., Jung, R. E., Güntürkün, O., & Genç, E. (2021). Interindividual differences in matrix reasoning are linked to functional connectivity between brain regions nominated by Parieto-Frontal Integration Theory. Intelligence, 87, 101545. http://dx.doi.org/10.1016/j.intell.2021.101545
Frischkorn, G. T., Schubert, A. L., & Hagemann, D. (2019). Processing speed, working memory, and executive functions: Independent or inter-related predictors of general intelligence. Intelligence, 75, 95–110. http://dx.doi.org/10.1016/j.intell.2019.05.003
Gignac, G. E. (2015). Raven's is not a pure measure of general intelligence: Implications for g factor theory and the brief measurement of g. Intelligence, 52, 71–79. http://dx.doi.org/10.1016/j.intell.2015.07.006
Güntürkün, O. (2005). The avian ‘prefrontal cortex’ and cognition. Current Opinion in Neurobiology, 15(6), 686–693. http://dx.doi.org/10.1016/j.conb.2005.10.003
Hedge, C., Powell, G., & Sumner, P. (2018). The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences. Behavior Research Methods, 50(3), 1166–1186. http://dx.doi.org/10.3758/s13428-017-0935-1
Hollard, V. D., & Delius, J. D. (1982). Rotational invariance in visual pattern recognition by pigeons and humans. Science, 218(4574), 804–806. http://dx.doi.org/10.1126/science.7134976
Isden, J., Panayi, C., Dingle, C., & Madden, J. (2013). Performance in cognitive and problem-solving tasks in male spotted bowerbirds does not correlate with mating success. Animal Behaviour, 86(4), 829–838. http://dx.doi.org/10.1016/j.anbehav.2013.07.024
Izquierdo, A., Brigman, J. L., Radke, A. K., Rudebeck, P. H., & Holmes, A. (2017). The neural basis of reversal learning: An updated perspective. Neuroscience, 345, 12–26. http://dx.doi.org/10.1016/j.neuroscience.2016.03.021
JASP Team (2020). JASP (Version 0.16.1). JASP Team. https://jasp-stats.org/
Jensen, A. R. (1992). Commentary: Vehicles of g. Psychological Science, 3(5), 275–279. http://dx.doi.org/10.1111/j.1467-9280.1992.tb00671.x
Jensen, A. R. (1998). The g factor: The science of mental ability (Vol. 648). Westport, CT: Praeger.
John, O. P., & Benet-Martinez, V. (2000). Measurement: reliability, construct validation, and scale construction. In H.T. Reis & C. M. Judd (Eds) Handbook of research methods in social and personality psychology (pp. 339–39). Cambridge University Press. http://dx.doi.org/10.1017/CBO9780511996481.023
Johnson, W., Bouchard Jr, T. J., Krueger, R. F., McGue, M., & Gottesman, I. I. (2004). Just one g: Consistent results from three test batteries. Intelligence, 32(1), 95–107. http://dx.doi.org/10.1016/S0160-2896(03)00062-X
Johnston, M., Anderson, C., & Colombo, M. (2017). Neural correlates of sample-coding and reward-coding in the delay activity of neurons in the entopallium and nidopallium caudolaterale of pigeons (Columba livia). Behavioural Brain Research, 317, 382–392. http://dx.doi.org/10.1016/j.bbr.2016.10.003
Jolliffe, I. T., & Cadima, J. (2016). Principal component analysis: a review and recent developments. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374(2065), 20150202. http://dx.doi.org/10.1098/rsta.2015.0202
Jung, R. E., & Haier, R. J. (2007). The Parieto-Frontal Integration Theory (P-FIT) of intelligence: converging neuroimaging evidence. Behavioral and Brain Sciences, 30(2), 135. http://dx.doi.org/10.1017/S0140525X07001185
Kangas, B. D., Berry, M. S., & Branch, M. N. (2011). On the development and mechanics of delayed matching-to‐sample performance. Journal of the experimental analysis of behavior, 95(2), 221–236. http://dx.doi.org/10.1901/jeab.2011.95-221
Karakuyu, D., Herold, C., Güntürkün, O., & Diekamp, B. (2007). Differential increase of extracellular dopamine and serotonin in the ‘prefrontal cortex’ and striatum of pigeons during working memory. European Journal of Neuroscience, 26(8), 2293–2302. http://dx.doi.org/10.1111/j.1460-9568.2007.05840.x
Katz, J. S., & Wright, A. A. (2006). Same/different abstract-concept learning by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 32(1), 80. http://dx.doi.org/10.1037/0097-7403.32.1.80
Kaufman, A. B., Reynolds, M. R., & Kaufman, A. S. (2019). The structure of ape (hominoidea) intelligence. Journal of Comparative Psychology, 133(1), 92–105. http://dx.doi.org/10.1037/com0000136
Kaufman, S. B., DeYoung, C. G., Gray, J. R., Brown, J., & Mackintosh, N. (2009). Associative learning predicts intelligence above and beyond working memory and processing speed. Intelligence, 37(4), 374–382. http://dx.doi.org/10.1016/j.intell.2009.03.004
Kolata, S., Light, K., Grossman, H. C., Hale, G., & Matzel, L. D. (2007). Selective attention is a primary determinant of the relationship between working memory and general learning ability in outbred mice. Learning & Memory, 14(1–2), 22–28. http://dx.doi.org/10.1101/lm.408507
Kolata, S., Light, K., Townsend, D. A., Hale, G., Grossman, H. C., & Matzel, L. D. (2005). Variations in working memory capacity predict individual differences in general learning abilities among genetically diverse mice. Neurobiology of Learning and Memory, 84(3), 241–246. http://dx.doi.org/10.1016/j.nlm.2005.07.006
Kovacs, K., & Conway, A. R. (2019). What is iq? life beyond “general intelligence”. Current Directions in Psychological Science, 28(2), 189–194. http://dx.doi.org/10.1177/0963721419827275
Lissek, S., Diekamp, B., & Güntürkün, O. (2002). Impaired learning of a color reversal task after NMDA receptor blockade in the pigeon (Columbia livia) associative forebrain (Neostriatum Caudolaterale). Behavioral Neuroscience, 116(4), 523. http://dx.doi.org/10.1037/0735-7044.116.4.523
Locurto, C., & Scanlon, C. (1998). Individual differences and a spatial learning factor in two strains of mice (Mus musculus). Journal of Comparative Psychology, 112(4), 344–352. http://dx.doi.org/10.1037/0735-7036.112.4.344
Locurto, C., Benoit, A., Crowley, C., & Miele, A. (2006). The structure of individual differences in batteries of rapid acquisition tasks in mice. Journal of Comparative Psychology, 120(4), 378. http://dx.doi.org/10.1037/0735-7036.120.4.378
Locurto, C., Fortin, E., & Sullivan, R. (2003). The structure of individual differences in heterogeneous stock mice across problem types and motivational systems. Genes, Brain and Behavior, 2(1), 40–55. http://dx.doi.org/10.1034/j.1601-183X.2003.00006.x
Major, J. T., Johnson, W., & Bouchard Jr, T. J. (2011). The dependability of the general factor of intelligence: Why small, single-factor models do not adequately represent g. Intelligence, 39(5), 418–433. http://dx.doi.org/10.1016/j.intell.2011.07.002
Matzel, L. D., Han, Y. R., Grossman, H., Karnik, M. S., Patel, D., Scott, N., Specht, S. M., & Gandhi, C. C. (2003). Individual differences in the expression of a “general” learning ability in mice. Journal of Neuroscience, 23(16), 6423–6433. http://dx.doi.org/10.1523/JNEUROSCI.23-16-06423.2003
Matzel, L. D., Sauce, B., & Wass, C. (2013). The architecture of intelligence: Converging evidence from studies of humans and animals. Current Directions in Psychological Science, 22(5), 342–348. http://dx.doi.org/10.1177/0963721413491764
Matzel, L. D., Townsend, D. A., Grossman, H., Han, Y. R., Hale, G., Zappulla, M., Light. K., & Kolata, S. (2006). Exploration in outbred mice covaries with general learning abilities irrespective of stress reactivity, emotionality, and physical attributes. Neurobiology of Learning and Memory, 86(2), 228–240. http://dx.doi.org/10.1016/j.nlm.2006.03.004
Mundfrom, D. J., Shaw, D. G., & Ke, T. L. (2005). Minimum sample size recommendations for conducting factor analyses. International Journal of Testing, 5(2), 159–168. http://dx.doi.org/10.1207/s15327574ijt0502_4
Neubauer, A. C. (1991). Intelligence and RT: A modified Hick paradigm and a new RT paradigm. Intelligence, 15(2), 175–192. http://dx.doi.org/10.1016/0160-2896(91)90029-D
O'Donoghue, E. M., Broschard, M. B., Freeman, J. H., & Wasserman, E. A. (2022). The Lords of the Rings: People and pigeons take different paths mastering the concentric-rings categorization task. Cognition, 218, 104920. http://dx.doi.org/10.1016/j.cognition.2021.104920
Peirce, J. W. (2007). PsychoPy—psychophysics software in Python. Journal of Neuroscience Methods, 162(1–2), 8–13. http://dx.doi.org/10.1016/j.jneumeth.2006.11.017
Pietschnig, J., Penke, L., Wicherts, J. M., Zeiler, M., & Voracek, M. (2015). Meta-analysis of associations between human brain volume and intelligence differences: How strong are they and what do they mean?. Neuroscience & Biobehavioral Reviews, 57, 411–432. http://dx.doi.org/10.1016/j.neubiorev.2015.09.017
Plenderleith, M. (1956). Discrimination learning and discrimination reversal learning in normal and feebleminded children. The Journal of Genetic Psychology, 88(1), 107–112. http://dx.doi.org/10.1080/00221325.1956.10532958
Quiroga, M. A., Diaz, A., Román, F. J., Privado, J., & Colom, R. (2019). Intelligence and video games: Beyond “brain-games”. Intelligence, 75, 85–94. http://dx.doi.org/10.1016/j.intell.2019.05.001
Rast, P., & Zimprich, D. (2009). Individual differences and reliability of paired associates learning in younger and older adults. Psychology and Aging, 24(4), 1001. http://dx.doi.org/10.1037/a0016138
Sackett, P. R., & Yang, H. (2000). Correction for range restriction: an expanded typology. Journal of Applied Psychology, 85(1), 112. http://dx.doi.org/10.1037/0021-9010.85.1.112
Salthouse, T. A. (1994). Aging associations: influence of speed on adult age differences in associative learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(6), 1486. http://dx.doi.org/10.1037/0278-7393.20.6.1486
Sauce, B., Bendrath, S., Herzfeld, M., Siegel, D., Style, C., Rab, S., Korabelnikov, J., & Matzel, L. D. (2018). The impact of environmental interventions among mouse siblings on the heritability and malleability of general cognitive ability. Philosophical Transactions of the Royal Society B: Biological Sciences, 373(1756), 20170289. http://dx.doi.org/10.1098/rstb.2017.0289
Schubiger, M. N., Fichtel, C., & Burkart, J. M. (2020). Validity of cognitive tests for non-human animals: pitfalls and prospects. Frontiers in Psychology, 11, 1835. http://dx.doi.org/10.3389/fpsyg.2020.01835
Shaw, R. C. (2017). Testing cognition in the wild: Factors affecting performance and individual consistency in two measures of avian cognition. Behavioural Processes, 134, 31–36. http://dx.doi.org/10.1016/j.beproc.2016.06.004
Shaw, R. C., & Schmelz, M. (2017). Cognitive test batteries in animal cognition research: evaluating the past, present and future of comparative psychometrics. Animal Cognition, 20(6), 1003–1018. http://dx.doi.org/10.1007/s10071-017-1135-1
Shaw, R. C., Boogert, N. J., Clayton, N. S., & Burns, K. C. (2015). Wild psychometrics: evidence for ‘general’ cognitive performance in wild New Zealand robins, Petroica longipes. Animal Behaviour, 109, 101–111. http://dx.doi.org/10.1016/j.anbehav.2015.08.001
Sheppard, L. D., & Vernon, P. A. (2008). Intelligence and speed of information-processing: A review of 50 years of research. Personality and Individual Differences, 44(3), 535–551. http://dx.doi.org/10.1016/j.paid.2007.09.015
Soha, J. A., Peters, S., Anderson, R. C., Searcy, W. A., & Nowicki, S. (2019). Performance on tests of cognitive ability is not repeatable across years in a songbird. Animal Behaviour, 158, 281–288. http://dx.doi.org/10.1016/j.anbehav.2019.09.020
Soravia, C., Ashton, B. J., Thornton, A., & Ridley, A. R. (2022). General cognitive performance declines with female age and is negatively related to fledging success in a wild bird. Proceedings of the Royal Society B, 289(1989), 20221748. http://dx.doi.org/10.1098/rspb.2022.1748
Spearman, C. (1904). "General Intelligence," objectively determined and measured. The American Journal of Psychology, 15(2), 201–292. http://dx.doi.org/10.2307/1412107
Spetch, M. L., Cheng, K., & MacDonald, S. E. (1996). Learning the configuration of a landmark array: I. Touch-screen studies with pigeons and humans. Journal of Comparative Psychology, 110(1), 55. http://dx.doi.org/10.1037/0735-7036.110.1.55
Stankov, L. (2017). Overemphasized “g”. Journal of Intelligence, 5(4), 33. http://dx.doi.org/10.3390/jintelligence5040033
Sternberg, R. J., & Gastel, J. (1989). Coping with novelty in human intelligence: An empirical investigation. Intelligence, 13(2), 187–197. http://dx.doi.org/10.1016/0160-2896(89)90016-0
Tamez, E., Myerson, J., & Hale, S. (2008). Learning, working memory, and intelligence revisited. Behavioural Processes, 78(2), 240–245. http://dx.doi.org/10.1016/j.beproc.2008.01.008
van den Heuvel, M. P., & Sporns, O. (2013). Network hubs in the human brain. Trends in Cognitive Sciences, 17(12), 683–696. http://dx.doi.org/10.1016/j.tics.2013.09.012
Van Ginkel, J. R., Kroonenberg, P. M., & Kiers, H. A. (2014). Missing data in principal component analysis of questionnaire data: a comparison of methods. Journal of Statistical Computation and Simulation, 84(11), 2298–2315. http://dx.doi.org/10.1080/00949655.2013.788654
van Horik, J. O., & Lea, S. E. (2017). Disentangling learning from knowing: Does associative learning ability underlie performances on cognitive test batteries?. The Behavioral and Brain Sciences, 40, e220. http://dx.doi.org/10.1017/S0140525X16001795
Vickrey, C., & Neuringer, A. (2000). Pigeon reaction time, Hick’s law, and intelligence. Psychonomic Bulletin & Review, 7(2), 284–291. http://dx.doi.org/10.3758/BF03212983
Völter, C. J., Tinklenberg, B., Call, J., & Seed, A. M. (2018). Comparative psychometrics: establishing what differs is central to understanding what evolves. Philosophical Transactions of the Royal Society B: Biological Sciences, 373(1756), 20170283. http://dx.doi.org/10.1098/rstb.2017.0283
Warne, R. T., & Burningham, C. (2019). Spearman’s g found in 31 non-Western nations: Strong evidence that g is a universal phenomenon. Psychological Bulletin, 145(3), 237–272. http://dx.doi.org/10.1037/bul0000184
Williams, B. A., & Pearlberg, S. L. (2006). Learning of three-term contingencies correlates with Raven scores, but not with measures of cognitive processing. Intelligence, 34(2), 177–191. http://dx.doi.org/10.1016/j.intell.2005.03.007
Wright, A. A. (1997). Concept learning and learning strategies. Psychological Science, 8(2), 119–123. http://dx.doi.org/10.1111/j.1467-9280.1997.tb00693.x
Wright, A. A., & Katz, J. S. (2006). Mechanisms of same/different concept learning in primates and avians. Behavioural processes, 72(3), 234–254. http://dx.doi.org/10.1016/j.beproc.2006.03.009
Wright, A. A., Kelly, D. M., & Katz, J. S. (2018). Comparing cognition by integrating concept learning, proactive interference, and list memory. Learning & Behavior, 46(2), 107–123. http://dx.doi.org/10.3758/s13420-018-0316-3
Wright, A. A., Santiago, H. C., Sands, S. F., Kendrick, D. F., & Cook, R. G. (1985). Memory processing of serial lists by pigeons, monkeys, and people. Science, 229(4710), 287–289. http://dx.doi.org/10.1126/science.9304205
Zanto, T. P., & Gazzaley, A. (2013). Fronto-parietal network: flexible hub of cognitive control. Trends in Cognitive Sciences, 17(12), 602–603. http://dx.doi.org/10.1016/j.tics.2013.10.001
Zentall, T. R. (2021). Macphail (1987) Revisited: pigeons have much cognitive behavior in common with humans. Frontiers in Psychology, 11, 618636. http://dx.doi.org/10.3389/fpsyg.2020.618636
Zentall, T. R., & Smith, A. P. (2016). Delayed matching-to-sample: A tool to assess memory and other cognitive processes in pigeons. Behavioural Processes, 123, 26–42. http://dx.doi.org/10.1016/j.beproc.2015.07.002

No competing interests reported.

Download PDF

Journal Publication

published 02 Nov, 2024

Read the published version in Animal Cognition →

Editorial decision: Revision requested
05 Nov, 2023
Reviews received at journal
19 Sep, 2023
Reviewers agreed at journal
06 Sep, 2023
Reviewers invited by journal
05 Sep, 2023
Editor assigned by journal
05 Sep, 2023
Submission checks completed at journal
30 Aug, 2023
First submitted to journal
25 Aug, 2023

You are reading this latest preprint version

Evidence for a Two Component Cognitive Structure in Pigeons (Columba livia)

Status:

Journal Publication

Version 1

Abstract

Introduction

Method

Subjects

Apparatus

Procedure

Data Analysis

Results

Individual Cognitive Tasks

Symbolic Match to Sample

Serial Reversal Learning

Delayed Match to Sample

Reaction Time

Cognitive Test Battery

Correlation matrix

Principal component analysis

Discussion

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1