First, the numbers are unofficial. Latvijas Fakti decided not to release them. This has some bizarre pre-history. A few weeks ago, Dzimtene party claimed that their true popularity is about 10 times higher than in Latvijas Fakti polls (11% instead of 1.2%) and asked the Latvian anti-corruption office to investigate Latvijas Fakti. Dzimtene is a moderately obscure party which recently attracted attention by nominating Vadim Putin (who has similar name but is not related to Russian president Vladimir Putin) as a candidate to Latvian parliament. That smelled like a publicity stunt and Dzimtene's claim about opinions polls being wrong has a similar smell. Some of people involved in Dzimtene have a history of such stunts and there is nothing surprising here. Nothing, except for the reaction of Latvijas Fakti. They said they are tired of people questioning their polls and decided not to release the most recent poll. (I wonder if that means that some more influential party is unhappy as well... Dzimtene is too obscure to merit such strong reaction.) Their decision was a pity because I respect Latvijas Fakti and their numbers. But, a few days later, newspapers got hold of the numbers and, now, everyone knows them.
Second, the big story is Jaunais Laiks (New Era) falling from the first place in July to the fourth place in August. Neatkariga Rita Avize (NRA), a newspaper hostile to New Era, credits the fall to their publication of transcripts of phone conversations between a businessmen who was being investigated by the tax service and senior New Era members. The businessmen was seeking for a help against the investigation and he was getting somewhat symphatetic response from New Era people. New Era Party has the same opinion about the reasons why they fall to the fourth place in the opinion poll, except that they call it as a "smear campaign by NRA".
I think they both may be reading too much into a small change in polls. New Era numbers have declined only by 1.3%. For a poll of 1000 people, that's 13 people and the probability of polling company accidentially choosing 13 less supporters of New Era this time is not the small. According to a calculation by this blogger, the probability of such event is 5-10%, meaning that New Era numbers would get underestimated by 1.3% once every 10-20 months. As long as there isn't another opinion poll confirming the change, it may just be random noise in data.