Flashcards
305 niga
Flashcards do conjunto (15)
- 1explain obersvation resampling? give a diagramobservation resampling treats (*x-y) as a pair and resamples from these to get new samples.
approprate where x cannot be controlled in advance, so have to account for sample variation in both x and y
graph x and y are random... circle plots - 2why is leverage an important feature when considering the influence of an outlierleverage increases as a explanatory value moves away from the explanatory mean.
-if an outlier has high leverage it will cause a great disruption of the linear model than a point with low or no leverage - 3what is a plot that can be used to detect outliers with high leverageplot residuals on the y axis, with leverage on the x axis, high residual with a high leverage suggests an outlier with strong leverage
- 4describe the features of anscombe's quarteta - simple linear relationship, linear appropriate
b- not a normal distribution (non linear) not linear appropriate
c- one outlier to a good fit linear appropriate
d - one outlier turning a terrible linear appropriate fit into a good fit - 5explain the difference between a confidence interval and a prediction interval for a new observationa confidence interval is a measure of how well the mean has been determined.
it gives an indication of the interval the mean falls into
*prediction intervals give an indication of where you would expect a new observation point to fall into ie takes into account of error - 6how can a 95% confidence interval be used to carry out a correlation hypothesis test for linear associationvalues inside the confidence interval are plausible for the correlation coefficient at x% confidence interval
if the values are close ( eg 0.98:1 or -1 :-0.98 ) then correlation occurs if they are wide then there is no linear association - 7how can the gradient coefficient be bootstrapped to show a linear association between two parameters?if the bootstrapped gradients are plotted using a histogram, the plot will give an indicator of linear association.
if the histogram contains a small amount of large adjacent values ?? then there is a fit
if not (ie values are spread ) there is no fit
if most are above zero but some not, insufficient info - 8what are the key assumptions when fitting a multiple linear regression model?-variables are normally distributed
- assumption of linear relationship between the explanatory and response variables
- variables have been measured without error ( reliable)
- assumption of homoscedasticity
-X is fixed ( weak exogeneity )
-errors uncorrelated - 9which plot of the residuals shows a good regression model ? what is wrong with the others refer to graphsa is best fit, though would be better if errors were zero
b shown a non constant variance
c shows a non constant variance
d exhibits a non-zero mean indicator ie there is a non linear distribution - 10how can a regression model be modified to capture a residual pattern that is non linearinclude the quadratic term as an explanatory variable
- 11sketch a conceptual understanding of the confidence and prediction intervals and explain why they are smallest near the centroidgraph with prediction and confidence interval and line best fit
smallest near the centroid as there is less uncertainty when predicting in the middle of the data. the intervals get larger as uncertainty gets larger - 12explain residual resampling and give a diagramresidual resampling treats X as fixed and takes a resample of y by adding the resampled residuals to the predictive y hat + ( resampled appropriate )
for experimental type studies where X are fixed in advanced
graph
x fixed
y varies - 13how does residual resampling mask an outliertthe outlying point drags down the orginal regression line , thereby masking its outlyinness
and stopping variation of residuals - 14what is suggested by a confidence interval of bootstrapped gradients that contains zero- rejection of the null hypothesis of correlation
- alternatively, accept hypothesis of no correlation - 15explain how the inverse transformation method can be used to obtain random numbers based on CDF include both diagrams and equations1) generate uniform distributed rand number, u(0,1)
2)obtain inverse cumulative distribution of random number
U=fx (x)
x=Fx^-1(U)
graph u1 x1 u2 x2
Informações
- Descrição
- FlashCards sobre 305 niga , criado por adam_longley em 23-06-2014.
- Criado por
- adam_longley
- Visualizações
- 19
- Avaliação