The dataset contains two files containing the number of visits made by each bird in each of the well types, out of the first 15 visits in each test (one file for Experiment 1 and another for Experiment 2). An additional file contains our handling time samples for each of the treatments used in experiment 1 (exposed, hidden and "wrapped").

During the summer of 2017 (August–September), and spring of 2018 (March–May) we captured house-sparrows in Kibbutz Hulda, and in the Meier Segal's Garden of Zoological Research of Tel-Aviv University, Israel, and housed them in large aviaries located at the zoological garden. Before the beginning of each experiment, the participating birds were moved from the aviaries to individual cages (size: 75 × 45 × 45 cm) positioned side by side within an outdoor aviary. The individual cages were separated from one another with a plastic partition that had a small window (10 × 10 cm) in its upper back corner, which allowed the birds to occasionally see their neighbors (to reduce isolation stress), but prevented them from seeing each other while foraging. Each cage contained a wooden foraging grid with 30 feeding wells (2.5 cm diameter, 1.8 cm deep and 8.5 cm distance between wells), a small water bowl, ad libitum supply of food (a mixture of seeds, mashed boiled eggs and cucumbers), and artificial foliage at the backside of the cage (for shelter). A narrow flap door at the front bottom of the cage was used to insert the foraging grid when necessary. The feeding wells of the foraging grids contained cups that were filled with colored sand (up to ~ 3/4 of their depth), each cup fitted into a well. Birds that participated in the experiments were randomly divided into experimental cohorts of 6 birds, used (trained and tested) in subsequent weeks (it was not feasible to train and test more than 6 birds simultaneously). Only sparrows in good condition (weighing > 26.5 g with no visible signs of medical problems) were selected for the experiment. The study was carried out under animal care permit no. 04-16-060 of Tel-Aviv University.

The learning tasks used for the experiments were based on training the sparrows to associate different sand colors with a food reward (millet seeds) positioned in three different settings: (1) the exposed setting – seeds were placed on top of the sand; (2) the hidden setting – seeds were hidden in the sand; (3) the wrapped setting – seeds were placed on top of the sand (like in setting (1)) but the well was covered with a fluted transparent plastic wrap which the sparrows had to tear to reach the seeds.

We used clearly distinctive sand colors and randomized their roles across individuals so that settings could not be confounded by colors. In addition, using Ocean Optics USB-4000 spectrometer, we confirmed the distinctive spectral peaks of the different sand colors, the lack of UV reflectance that can bias the results, and the similarity of the spectral peaks of each color with and without the transparent wrap. Millet seeds, which are light yellow in color, were clearly visible on the background of all types of colored sand in the exposed and wrapped conditions.

We ran two experiments with different sets of birds (Experiment 1 and Experiment 2). Experiment 1 tested the effect of all three settings using the same amount of reward (3 millet seeds). In Experiment 2, we varied the amount of reward and used only the exposed and hidden settings. Both experiments included a pre-training stage (not shown in the figure) and then two parts, A and B, each comprised of a training phase and two subsequent tests (see Fig. 1). We first describe here the general structure of the two experiments, and then, in the two sub-sections below, each experiment and its rationale will be described in more detail.

The schedules of Experiments 1 and 2 were similar: on the first day, 6 sparrows were caught from the sparrows’ aviary and housed in individual cages (a cohort of 6 birds as mentioned above). On the second day (pre-training), we allowed the birds to learn how to extract millet seeds from wells in each of the involved settings (exposed, hidden and wrapped in Experiment 1; exposed and hidden in Experiment 2), using natural sea sand. The use of natural sand during this pre-training stage allowed the birds to learn the physical properties of the task but without yet associating different sand colors with different probabilities or quantities of food. On the third day, we carried out the first part of the experiments (Part A). The second part (Part B) was carried out on day 4 in Experiment 1 and on day 5 in Experiment 2 (this difference in schedule was for technical reason and should not affect the results as these were two different experiments with different sets of birds). Each part of the experiments (A or B) started with 2.5 h of food deprivation, followed by four training sessions (15 min each, no breaks between sessions), and ended with two 30 min test sessions (see details below).

In each of the four training phases of experiments 1 and 2, the birds were presented with two settings (e.g. hidden and exposed) associated with two distinct sand colors, the roles of which were counterbalanced between birds. A training phase was comprised of 4 sessions, in which the birds were presented each time with one of the two settings, in alternating sessions (see Fig. 1). In a given session, the birds could forage on a grid offering 15 feeding wells that were identical in setting, sand color, and reward (see Fig. 1; in these sessions the feeding wells were only available on half of the grid, and the other half was covered with a dark brown Polycarbonate sheet – for clarity, these covered parts are not shown in the figure). The reason for presenting different settings in different training sessions rather than mixing them within a session was the need to prevent primacy or recency effects (the tendency to best recall either the first or last items in a series, respectively). These effects may occur in a mixed presentation, where the sparrows are likely to approach the visible seeds before searching for the hidden ones. Unless otherwise specified (see article for full details), the setting used first in the training phase was chosen randomly. Each training session was designed to last 15 min but in cases where some of the birds did not explore the grid during this time period, 5 min were added to the session, as well as to the following session (to maintain equal exposure to the two settings). When the extra time was needed, it was also added to all the birds in the cohort.

The test sessions were carried out immediately after the four training sessions and lasted 30 min each. During the test sessions, the birds were presented with a foraging grid offering 15 feeding wells of each color in a random spatial distribution (30 wells in total). Within each test, all feeding wells were identical in setting and reward and differed only in sand color. Thus, the goal of the tests was to reveal how color preference exhibited in a particular setting is affected by the information previously learned about this color under different settings. Note that in tests of the exposed and the wrapped settings, food was also provided, as these settings must include visible seeds. In contrast, in tests of the hidden setting, no seeds were provided as this setting looks initially the same with and without seeds. This method of testing color preference after training with hidden seeds minimizes further learning during the test and has been proven reliable and informative in previous studies.

Training and test sessions were video-recorded and analyzed using a Python-based software (Perspective Birdy 1.0. https://arnonlotem.weebly.com/technical-tools–code.html) developed in our lab by Y. Perry and E. Shellef. Each visit to a well that included at least one peck, was defined as a foraging step and classified according to well type. Repeated pecks at the same well were distinguished as different foraging steps only if they were separated by a peck in another well. Sparrows’ preferences in the tests were analyzed based on their first 15 steps (excluding individuals who did not perform at least 6 foraging steps). We chose this number of steps a priori, because it is powerful enough to allow binomial analysis of individual preferences but also small enough to minimize the effects of extinction (in hidden tests that offer no food) or food depletion (in wrapped and exposed setting tests where there are 15 wells of each color offering food). Moreover, during the first 15 steps on the grid there are still alternative intact feeding wells of both color types that are immediately available to the birds, which makes differences in search time relatively negligible.

Training and test sessions were video-recorded and analyzed using a Python-based software (Perspective Birdy 1.0. https://arnonlotem.weebly.com/technical-tools–code.html) developed in our lab by Y. Perry and E. Shellef.