Keep Cause and Effect Clear

It's difficult to reason about a test when the cause is hidden far away from the effect.

Bad

class TallyTest : public ::testing::Test {
protected:
  void SetUp() override {
    tally_.Increment("key1", 8);
    tally_.Increment("key2", 100);
    tally_.Increment("key1", 0);
    tally_.Increment("key1", 1);
  }

  Tally tally_;
}
// 200 lines of code

TEST_F(TallyTest, IncrementExistingKey) {
  EXPECT_EQ(9, tally_.Get("key1"));
}

Write tests where the effects immediately follow the causes.

Good

class TallyTest : public ::testing::Test {
protected:
  Tally tally_;
}

TEST_F(TallyTest, NewKey) {
  tally_.Increment("key", 100);
  EXPECT_EQ(100, tally_.Get("key"));
}

TEST_F(TallyTest, ExistingKey) {
  tally_.Increment("key", 8);
  tally_.Increment("key", 1);
  EXPECT_EQ(9, tally_.Get("key"));
}

TEST_F(TallyTest, IncrementByZeroDoesNothing) {
  tally_.Increment("key", 8);
  tally_.Increment("key", 0);
  EXPECT_EQ(8, tally_.Get("key"));
}

It may require a bit more code, but it's easier to read and maintain.