d*******1 发帖数: 854 | 1 我有一个化学实验, 取样是在某一天的某个plate上的的某一个well (column X row)
, 我有如下的linear model
y= U+ day+ plate(day)+ column + row + e.
我已经有个一个sample dataset, 做好了ANOVA, 有了ANOVA TABLE, 各种因子的 mean
square已经算好, 我现在想根据这些variance demoposition的结果进行一个大的
simulation去模拟真实的数据, 请问如何做 SIMULATION 呢 (已经知道分布是normal
)? sas 或者R 均可。谢谢 | d*******1 发帖数: 854 | 2 nobody? 自顶一下。
row)
mean
normal
【在 d*******1 的大作中提到】 : 我有一个化学实验, 取样是在某一天的某个plate上的的某一个well (column X row) : , 我有如下的linear model : y= U+ day+ plate(day)+ column + row + e. : 我已经有个一个sample dataset, 做好了ANOVA, 有了ANOVA TABLE, 各种因子的 mean : square已经算好, 我现在想根据这些variance demoposition的结果进行一个大的 : simulation去模拟真实的数据, 请问如何做 SIMULATION 呢 (已经知道分布是normal : )? sas 或者R 均可。谢谢
| j******o 发帖数: 127 | 3 没有这方面的经验。你是说你的model中day, plate(day), column and row都是随机变
量?你知道所有变量的distribution? 你也知道所有的parameters?然后你想simulate
the sum?对吗? | d*******1 发帖数: 854 | 4 they are not random variable. they are fixed effect (can be random effect as
well...). I can estimate the mean sum of square for these using sample data
(mean sum of square for error (MSE) is unbiased estimator of random
variance). I wonder how to randomly generate these assay data that mimic the
reality (data collected on various dates and plates and on various
locations of plates) based on the known source of variance attributed by
these factors (again, they are estimated from sample data)....
【在 j******o 的大作中提到】 : 没有这方面的经验。你是说你的model中day, plate(day), column and row都是随机变 : 量?你知道所有变量的distribution? 你也知道所有的parameters?然后你想simulate : the sum?对吗?
| z**k 发帖数: 378 | | j******o 发帖数: 127 | 6 So the residual is only random variable in your model. With the data you
have, you can get residuals using your model. Then the mean and std of
residuals can be calculated. thus, sampling from normal distribution with
parameter you estimated could be the place that your simulation starts from.
as
data
the
【在 d*******1 的大作中提到】 : they are not random variable. they are fixed effect (can be random effect as : well...). I can estimate the mean sum of square for these using sample data : (mean sum of square for error (MSE) is unbiased estimator of random : variance). I wonder how to randomly generate these assay data that mimic the : reality (data collected on various dates and plates and on various : locations of plates) based on the known source of variance attributed by : these factors (again, they are estimated from sample data)....
| d*******1 发帖数: 854 | 7 thank you. you are right that, in this case, residual is the only random
variance. if we know the sample mean is u and we can estimate fixed effect
for day1 is d1 and fixed effect for plate 1 is p1 and residual variance is
sigma, then we can just simulate the experimental data collected on day1 and
plate 1 by generating random number from N(u+d1+p1, sigma).
however, I now realize that it would be more appropriate to treat day and
plate as RANDOM effect, so I run mixed model on sample data and ob
【在 j******o 的大作中提到】 : So the residual is only random variable in your model. With the data you : have, you can get residuals using your model. Then the mean and std of : residuals can be calculated. thus, sampling from normal distribution with : parameter you estimated could be the place that your simulation starts from. : : as : data : the
|
|