**Within this project we share:** - Experimental task - The raw data file - R analysis script allowing to reproduce all the analyses reported in the paper. Order of analyses in the R script follows the order of reporting the data in the paper. **Experimental Task** Scripts running in DMDX software along with full documentation. For details see pdf file inside the compressed folder. **Scripts** Please note that due to implementation of bootstrap procedures, running the script might take some hours. Additional R scripts allow: - checking that results remain unchanged when standardized numerical distance and size effect slopes are considered. - conducting analysis for the ratio effect (as reported in Supplementary Material) By uncommenting one line directly after loading the raw data, it is possible to reproduce the analysis considering male participants only (As reported in Supplementary Material). All R scripts are extensively commented to increase readability. Data is stored in the tab-delimited txt file, so that it can be open in virtually spreadsheet and data analysis program. **Organization of the data file** The data are stored in the long format (i.e., one row corresponds to one trial). Data from practice sessions are excluded. The order of rows follows order of trials presented to given participant. Variables: part.code - unique identifier of a participant. Please note that the codes are the same as codes from parity judgment task performed by these participants just before the magnitude classification task reported here. These data are shared in another OSF project (https://osf.io/syeuw/ for reference sample and https://osf.io/bvyxn/ for mathematician, engineer and social science groups). Participant numbers are completely arbitrary, and are used solely to differentiate one participant form another. group - group considered in the analysis. Values: M = mathematicians, E = engineers, S = social sciences & humanities, R = reference sample. This information is replicated across all rows corresponding to a given participant. age - participant’s age in years (integers only). This information is replicated across all rows corresponding to a given participant. gender - participant’s gender: values F and M. This information is replicated across all rows corresponding to a given participant. handedness.coded - Handednees coded into left- and right-handers. In case of M, E, and S groups being a right-hander was an inclusion criterium. For the reference sample, handedness was recoded from LQ (see below) so that values above 0 were coded as R and values below 0, as L. This information is replicated across all rows corresponding to a given participant. lq (laterality quointent) - score from Oldfield's handedness questionnaire. Only available for the R group. Values from -100 (full left hand dominance) to 100 (full right hand dominance). This information is replicated across all rows corresponding to a given participant. block.order - order of blocks (i.e., response to hand assignments) administered to a given participant. Two values are possible for this variable: start.smaller.right and start.smaller.left. This information is replicated across all rows corresponding to a given participant. block.no - block number according to order of administration: Possible values: 1 = first block in the experiment, 2 = second block in the experiment. series - randomising of trials was built in such a way that all possible numbers were presented in random order, and only when all have been presented the same set was presented again in a randomised order. Therefore, on could differentiate series in which all numbers appeared. For each block in each participant numbered continuously from 1 to number corresponding to number of repetitions of each number per block. trial.no - trial number. Continuous numbering from 1 to 480 (number of trials / participant). number - number presented to the participant in a given trial. Possible values: 1, 2, 3, 4, 6, 7, 8, 9. resp.side - response, on which the correct response button was located. Please note that this indicates location of the correct response, not the actual key that was pressed (this differ in case of incorrect responses). Possible values: left and right. rt - reaction time in milliseconds. correct.experimental - filtering variable excluding incorrect responses and trials in which participants did not respond at all. Possible values: 0 = incorrect response / no response, 1 = correct response. filter.correct.exp.antic - filtering variable excluding anticipations < 200 milliseconds. Its value also equals zero for incorrect responses (indicated by correct.experimental above). Possible values: 0 = invalid trial, 1 = valid trial. filter.seq - sequential filter as described in the paper. Outlier reaction times (deviating by more than 3SD from individual mean for each participant) were sequentially trimmed. Its value also equals zero for incorrect responses and anticipation (indicated by filter.correct.exp.antic above). Possible values: 0 = invalid trial, 1 = valid trial.