The document discusses how the normal approximation for the binomial distribution depends on the number of trials and the probability of success on each trial. It notes that with 1000 trials and a probability of success of 1/1000, the distribution is not close to normal. However, with 10,000 trials from the same probability distribution, the distribution begins to approach normal. It advises checking that expected value plus or minus 3 standard errors does not produce impossible values before applying the normal approximation to a binomial distribution.
The document discusses how the normal approximation for the binomial distribution depends on the number of trials and the probability of success on each trial. It notes that with 1000 trials and a probability of success of 1/1000, the distribution is not close to normal. However, with 10,000 trials from the same probability distribution, the distribution begins to approach normal. It advises checking that expected value plus or minus 3 standard errors does not produce impossible values before applying the normal approximation to a binomial distribution.
The document discusses how the normal approximation for the binomial distribution depends on the number of trials and the probability of success on each trial. It notes that with 1000 trials and a probability of success of 1/1000, the distribution is not close to normal. However, with 10,000 trials from the same probability distribution, the distribution begins to approach normal. It advises checking that expected value plus or minus 3 standard errors does not produce impossible values before applying the normal approximation to a binomial distribution.
p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 How large is large? 1000 independent, repeated S/F trials p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 How large is large? 1000 independent, repeated S/F trials p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 How large is large? 1000 independent, repeated S/F trials p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 How large is large? 1000 independent, repeated S/F trials p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 How large is large? 1000 independent, repeated S/F trials p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 How large is large? 1000 independent, repeated S/F trials p = 1/1000 chance of success on a single trial Probability distribution of the number of successes binomial n = 1000, p = 1/1000 = 0.001 Sum of 1000 draws at random with replacement from 1 1 999 0 expected value of the number of successes = 1000 0.001 = 1 SE of the number of successes =
1000 0.001 0.999 = 0.9995
Not normal! Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 1 / 1 The probability histogram 1000 trials, p = 1/1000 number of successes p r o b a b i l i t y 0 1 2 3 4 5 6 0 . 0 0 . 1 0 . 2 0 . 3 Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 2 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 When the box is far from symmetric 1000 seems large but 1 1 999 0 is such an uneven distribution that 1000 draws iare not enough for the CLT to kick in What if we made 10,000 draws from the same box? expected value of the number of successes = 10, 000 0.001 = 10 SE of the number of successes =
10, 000 0.001 0.999 = 3.16
Normal? Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 3 / 1 Approaching normal 10,000 trials, p = 1/1000 number of successes p r o b a b i l i t y 0 5 10 15 20 25 30 0 . 0 0 0 . 0 4 0 . 0 8 0 . 1 2 Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 4 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1 Bad boxes It takes lots of draws for the probability histogram for the sum to start looking normal if the contents of the box are far from symmetric, for example skewed have a histogram that has gaps Practical advice: Calculate expected value 3SE and check whether you are hitting impossible values for the variable. If you are, then dont use the normal approximation. If you are not, then go ahead, but be aware that it still might not be great . . . Ani Adhikari and Philip Stark Statistics 2.2X Lecture 4.5 5 / 1