Greetings travelers! Since the Lucky Friends feature was released in April 2019, Silph researchers have been busy interacting with their friends through raids, trades, battles, and Gift opening to narrow down the lucky rate. With the likelihood of a lucky trade between two recently caught Pokémon sitting at 5.02%, the guaranteed lucky trade with a Lucky Friend is invaluable for collecting rarer Lucky Pokémon. In the pursuit of answers, 87 researchers have completed an impressive 56,780 friendship interactions. Let’s explore their findings!
- We found an overall rate of 1.17% for becoming Lucky Friends.
- There are only minor differences in lucky rates between interaction types.
- Historic Lucky Friend interactions have no influence on Lucky Friend outcome.
Background & Methodology
During data collection, a strict protocol was followed. Researchers ensured that data was only collected from Best Friends and that no prior interactions with each friend had already taken place that day. In the example below, the blue halos around the first two avatars indicate that interactions have already occurred that day. For Gift interactions, the Silph researcher was always the one to initiate the friendship interaction. All data used in this study was collected between April 12, 2019 and May 22, 2021, and interactions during the Lunar and Valentine’s events were excluded.
Even with these measures in place, we noticed a fair amount of reporting bias in our data collection. Reporting bias comes from the tendency to report memorable events more often than uninteresting ones; it often occurs when gathering data on infrequent events. In our case, this appeared as a tendency for researchers to over-report rare lucky interactions and under-report the more common non-lucky interactions. This occurred disproportionally among our less experienced researchers. To help reduce this reporting bias, we only considered data collected by the 87 experienced Silph Researchers who had contributed at least 100 data points to the study. While we suspect that the rates presented in this article still tend high, we are reasonably confident they give a close approximation of the true values.
Lucky Friends Rate by Interaction Type
Each interaction type was evaluated independently as well as collectively, and the table below shows the lucky outcomes for each interaction type. We excluded EX Raid invite interactions, as very few were reported and EX Raids are not currently available.
|Interaction Type||Lucky||Not Lucky
||Lucky Rate||95% Confidence Interval
|Gift opened||464||40,981||1.12%||[1.02 – 1.23%]|
|Trainer Battle||90||4,913||1.80%||[1.46 – 2.22%]|
|Pokémon Trade||29||3,049||0.94%||[0.64 – 1.37%]|
|Gym battle||8||985||0.81%||[0.38 – 1.65%]|
|Raid battle||71||5,932||1.18%||[0.93 – 1.50%]|
At first glance, most interaction types appear to show a similar rate of becoming lucky, with Trainer Battles being the outlier. Using a chi-squared test of independence, we found a significant association between the type of interaction and the interaction’s likelihood of resulting in a Lucky Friend (χ² (4, N = 56,522) = 20.5, p < 0.001, Cramér’s V = 0.019). Pairwise comparisons showed that the lucky rates for Trainer Battles were significantly higher than those from opening Gifts (χ²(1, N = 46,448) = 17.5 p < 0.001, Cramér’s V = 0.020) and trading Pokémon (χ²(1, N = 8,081) = 9.6, p = 0.002, Cramér’s V = 0.035). However, Cramér’s V effect size values (all values < 0.05) suggest little to no meaningful significance. With this in mind, let’s consider a few factors before drawing conclusions from this data.
In large datasets such as this one, even very small differences can be statistically significant. Tiny effect sizes such as those observed in this study could be the result of underlying variables on which we can only speculate. Perhaps Niantic coded two different lucky rates in an effort to encourage PvP in the time before the introduction of GO Battle League. Other possibilities include: a mistake with RNG seeding that happens only in Trainer Battles; contamination from reporting bias remaining in our data set that we were unable to completely remove; or the existence of a tiny partner-specific variation effect that we were unable to identify. Despite the statistical significance of the higher lucky rate from Trainer Battles, the very small effect size means identifying the cause is difficult.
Overall Lucky Friend Rate
Combining the data from all interaction types, our researchers experienced a lucky interaction rate of approximately 1.17% with a 99% confidence interval (CI) of 1.06 – 1.29%. On a practical level, the +0.68% increase in lucky rate for Trainer Battles is unlikely to be worth the extra time and coordination required to complete a Trainer Battle versus sending a Gift. If we exclude Trainer Battles from the data, the overall rate for the remaining interaction types falls to 1.12% (99% CI 1.00 – 1.23%, N = 51,776).
During the 2020 and 2021 Lunar events, Niantic promoted a boosted Lucky Friend rate. Silph Research observations showed the boosted rate across the two Lunar New Year events to be 3.08% (99% CI 2.36 – 4.01%, N = 3,053), drastically higher than the standard Lucky Friend rate.
Factors With No Influence
During the study, we confirmed that several factors have no influence on triggering Lucky Friend status. Factors tested include: Total All-time Lucky Friend Count (the total number of times the traveler has ever become Lucky Friends); Current Lucky Count (the number of Lucky Friends on the Friend List at the time of the interaction); and Lucky Count with Partner (the number of times the traveler has been Lucky Friends with the friend in question). We evaluated each researcher’s lucky interactions using a multiple logistic regression model and found they had no impact on the likelihood for another lucky interaction.¹
In addition, we looked at interactions which took place between partners who were in the same time zone versus different time zones. We found no association between time zone and lucky outcome (χ²(1, N = 56,780) = 0.49, p = 0.48).
Because this research continued for over two years, we had to consider the possibility of the Lucky Friend rate being silently changed during the study. As shown in the graph below, we broke down the lucky rate along with a 95% CI for each month. We found no association between month and the frequency of Lucky Friends (χ²(25, N = 56,780) = 28.5, p = 0.28). Lunar New Year & Valentine’s event data was excluded from this test because of their boosted rates.
For the average player, simply maximizing the number of Best Friend interactions will be the best way to gain Lucky Friends. If your play style easily allows the coordination of Trainer Battles, then you may want to take advantage of the slightly better lucky rate from unlimited daily Trainer Battles. For most trainers, the additional time required won’t justify the extra lucky opportunities. A trainer with 30 Best Friends who opens Gifts from all of them every day will spend about 36.5 hours during one year and reach Lucky Friends status approximately 123 times. If the same trainer does Trainer Battles as their first interaction with 5 Best Friends and opens Gifts from the rest, they’ll gain approximately 135 Lucky Friends but spend almost twice as long doing so (60.8 hours per year).² No matter how you choose to interact with your Best Friends, find the interaction method that works for you!
Although we have learned so much about the Lucky Friends feature, we may have more questions to answer down the road. For example: will the previously hinted at return of EX raids provide more opportunities to trigger Lucky Friends through EX raid invites? Until next time, travelers!
Author: Lead Researcher Belle
Analyst: Scientist CaroKann
Project Leaders: Scientists skyeofthetyger and IvoryTinkler1 and Lead Researchers alohanico and EagleTJM
Graphics Artists: Scientists CaroKann and WoodWoseWulf
Editors: Scientist Cham1nade and Lead Researcher JinianD
Data Collection: These 14 researchers were the most committed to collecting Lucky Friends:
¹ For the multiple logistic regression model, we estimated the log likelihood of a lucky outcome as predicted by Total All-time Lucky Friend Count, Current Lucky Count, and Lucky Count with Partner. Data where any of these values were unknown were excluded from this test (N = 50,366). A Wald test on the overall regression coefficients found the combined effect of the predictor variables to be insignificant (χ²(3) = 2.5, p = 0.48). Individually, Total Lucky Friends (z = 0.84, p = 0.40), Current Lucky Friends (z = -1.57, p = 0.12) and Lucky Count with Partner (z = -0.09, p = 0.93) were all poor predictors.
² We estimate that opening a Gift takes roughly twelve seconds. Finishing a Trainer Battle takes sixty seconds when both trainers enter combat immediately with prepared teams, and one team contains the lowest possible CP Pokémon to ensure a quick loss.