Hi all, firstly, I would like to thank you for the overwhelming response for the survey I created one day ago. I did not expect to be able to get that much amount of data in such a short period of time.

​

I. RAW DATA

​

The raw data have been exported in pdf formats and can be found here.

Let me have a few words about the raw data. Since I use the free account for survey monkey, the site stops giving results if from the response 101th onwards unless I pay subscription fee which isnt cheap. To work around this, I had to create multiple surveys and kept updating the hyperlink in the original post so the responses could be split and stored properly. Overall, there should be a total of 7 surveys.

​

II. METHODS

​

Firstly, as you can see, there are a few responses that claimed to summon more than 8 rainbows within one single 10 1 pulls. I did not include these responses into my analysis as these entries are clearly trolls. Please bear in mind that, the decision to not include these entries was made based on actual statistical tests which I shall not elaborate here (but if you are interested in how I did the tests, private message me and we can have a chat).

​

To calculate the expected number of CG Charlotte (CGC) and rainbows for each step, I use the following formula:

1. For step 1, 2 and 4: Expected CGC = total number of 10 1 pulls x (9 x 0.015 2 x 0.056) and Expected number of rainbows = total number of 10 1 pulls x (9 x 0.05 2 x 0.07)
2. For step 3: Expected CGC = total number of 10 1 pulls x (8 x 0.015 2 x 0.056 0.086) and Expected number of rainbows = total number of 10 1 pulls x (8 x 0.05 2 x 0.07 1)
3. For step 5: Expected CGC = total number of 10 1 pulls x (8 x 0.015 2 x 0.056 1) and Expected number of rainbows = total number of 10 1 pulls x (8 x 0.05 2 x 0.07 1)

After obtaining the observed and expected values, I used Chi-square test to test if the advertised rainbow rates are accurate. The alpha value I used to reject the null hypothesis was 0.01.

​

III. RESULTS

​

​

IV. DISCUSSIONS

​

For discussion, there are two main concerns as raised by various commentators in the original post:

1. This method of community pooling is solely based on trust and, therefore, subjected to falsification (e.g troll, people who purposely put bad results to make Gumi look bad, Gumi employees may try to put positive results to balance it out)
2. The sampled pool of players can be biased as people with bad pull results tend to report more than those who do better

While I totally agree upon the validity of such concerns, I want to emphasize the following points about the collected data

1. Survey monkey has in-built system to prevent one person from voting multiple times
2. The data for each sets of 80-100 responses as seen in the raw data show consistent trend and distribution. These subsets of 80-100 responses to the survey were collected over time and the transition from one set to another was not announced.
3. The results are mixed, not totally favoring any conclusions.

I also want to emphasize that this work was not professionally planned and executed so it could not be used as evidence for legal action, not anywhere near. Indeed, I hope through this small project, the community has some reference about what kind of rates they are up to and make their own judgement about what to do for the future. Feel free to comment and suggests what we should continue to improve transparency in FFBE.

​

TLDR;

- The overall on banner rates and rainbow rates for CG Charlotte Step up banners do not seem to be as advertised

- The on-banner rates for steps 1 and 5 are as advertised

- The rainbow rates for all steps and on-banner rates for steps 2, 3 and 4 do not seem to be as advertised

​

Note: results and analyses for 2.5k pull will come on later date.