Tag Info

16

Let's have a look at two vectors, the first being 2 6 2 6 2 6 2 6 2 6 2 6 and the second vector being 6 2 6 2 6 2 6 2 6 2 6 2 Calculating the Pearson correlation you'll get cor(a,b) [1] -1 However if you take the average of successive pairs for values both vectors are identical. Identical vectors have correlation 1. 4 4 4 4 4 4 This ...

14

The true distinction between data, is whether there exists, or not, a natural ordering of them that corresponds to real-world structures, and is relevant to the issue at hand. Of course, the clearest (and indisputable) "natural ordering" is that of time, and hence the usual dichotomy "cross-sectional / time series". But as pointed out in the comments, we ...

13

The answer here is pretty straight forward: Both pooled cross sectional data and pure panel data collect data over time (this can range from 2 time periods to any large number). The key difference between the two is the "units" we follow. I am defining units as households, countries, or whatever we are collecting data on. In pooled cross section, we will ...

10

Averaging can be attractive or convenient. It can also be a source of deception, at worst deceit, so tread carefully even when there is a clear rationale for averaging. Here is a situation it which it is not a good idea. Consider that by careful definition of groups you (usually) could reduce your data to two summary points each distinct on the two ...

6

To quote John Tukey: The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data. That is, there does not exist a statistical method that is a simple as causal_effect y x, int_validity="high" ext_validity="high" If any one claims to have something like this, it'...

4

fellow student here, so I hope a real professional can come in and give his advice. This advice is just from experience, and I would love to be corrected. Step 1: Decide what you are looking for. I assume you have already done this and not specified it in your question. For example, are you going to look at the density of cone cells, or maybe the optic ...

4

The basic fixed effect model is something like: $$y_{ij} = \boldsymbol{\beta} \cdot \mathbf{x}_{ij} + u_j + \epsilon_{ij}$$ Typically, either $j$ or $i$ is indexing over time, but there's no reason it can't be anything else. The math doesn't care. $i$ could index over students and $j$ could index classroom.

4

When you transform the data as you describe, the problem is that the rows in your data matrix no longer represent independent samples. While users may plausibly be assumed to be independent samples, time points for a given user are very likely to be dependent on previous time points. So this would violate the assumption that samples in your training and test ...

3

Depending on the structure of your dataset, it might even be possible to cluster in two dimensions, i.e. house and firm level. It depends on whether the house and firm level are nested or not. If they are, ignore what I say and go to the very good answer of Dimitriy. Petersen (2008) gives the theoretical justification for clustering on both time and firm ...

3

It's hard to answer your question precisely since it is not at all clear what you are doing. However, there are some general principles courtesy of Cameron and Miller's JHR paper. There's no formal test that will tell you at which level to cluster. If you think that the regressors or the errors are likely to be uncorrelated within a potential group, then ...

3

In a cross-sectional study, you are almost always getting prevalence data, so as a first step, you could consider these prevalence ratios. But, it sounds like you are modeling the number of symptoms as your outcome based on some covariates, using Poisson or negative binomial models. So you have something like this model: $log(Symptom Count|Gender) =\beta_{... 3 In this paper by Seawright (2009) several approaches how to analyze pseudo-panel data are disscussed, including: Cohort averaging (Deaton 1985; Verbeek & Nijman 1992) Two-Stage Auxiliary Instrumental Variables (2SAIV) (Franklin 1989) Multiple imputation (Gelman/King/Liu 1999) Matching (Rubin 2006) Three of the approaches are assesed against a true ... 3 I don't know Stata, but I'm going to guess that the equivalent is something like this: library(lme4) library(splines) ## a 'base' package glmer(y~x1+x2+x3+ns(time,3)+(1|country),family=binomial,data=dat) where splines::ns(time,3) indicates a natural cubic spline with 3 knots (knot placement is chosen automatically: see ?ns). If you have access to Stata ... 3 In short, there is no such thing as Multivariate Time Series data. The only classic data types out there are: Cross Sections, Time Series, Pooled Cross Sections, and Panel data. Panel data is multidimensional. Time series is one-dimensional. Time Series data is a type of panel data. Daily closing prices for last one year for 1 company is a Time Series ... 3 Beta regression as proposed by Ferrari & Cribari-Neto (2004, Journal of Applied Statistics) can be used to model the parameters of a beta regression by linear predictors. Specifically, the beta distribution is then not characterized by two shape parameters$p$and$q$but instead by mean$\mu$and precision parameter$\phi$where$\mu = p/(p + q)$and$\...

3

Panel data and longitudinal data are the same thing - the former terminology is more common in econometrics. Cross-sectional data are all collected at the same time. Therefore you have longitudinal / panel data, though I would rather call your study a type of "analysis of change"

3

I believe the explanation is this (haven't read the whole paper): The "Statistical Analysis" section starts with "To represent the general Korean population with minimal bias, sampling weights were applied to account for the complex sampling." The Table represents estimators computed from the sample; you can see this from the fact that ...

2

You partially answer your own question by asking for "longitudinal" changes. Cross-section data are called because they take a snap shot in time, literally a cross-section sliced out of a time-evolving society with its many relationship. Therefore, the best inference you can hope to do is under the assumption that whatever it is you are studying is time-...

2

It appears you have estimated a relation $$\ln Y = a_0 + a_1(X_1/Z_1) + a_2 \ln X_2$$ so $$a_2 = \frac {\partial \ln Y}{\partial \ln X_2}$$ By treating partial differentials as quantities (I won't tell if you don't) We have $$\frac {\partial \ln Y}{\partial \ln X_2} = \frac {\partial \ln Y}{\partial \ln X_2}\cdot \frac {1}{\partial X_2/\partial X_2} =... 2 It depends who your intended audience is and how strongly they respond to terminology. In principle you can redefine a word to mean whatever you want it mean, as long as you are clear about it. So in some situation you would be fine with using the term incidence rate ratio and add a footnote describing the difference between the way you used the term and the ... 2 Guess what: with a panel so short in its time dimension, non-stationarity of the whole process doesn't matter, because it is "unreachable". To illustrate: Consider the following simulated random walk with drift:$$x_t = 0.07 + x_{t-1} + u_t with $u_t \sim N(0,1)$ and independent from past and current regressors. An $800$-period graph looks like The two ...

2

From a theoretical point of view you already excluded this option and I think that is the correct way. Especially propensity score matching is sensitive to the matching covariates (see for example Smith and Todd, 2005). That's the most promising approach so I'll expand on this below. Depends what you want. It's not impossible to estimate treatment effects ...

2

The problem is coming from the fact that Fama-MacBeth standard errors do not include corrections regarding the fact that $\beta$'s are estimated (from the first stage TS regression). At least what you can do is compute the Shanken correction factors and check that the corrections in this case are indeed large. And that's why you are having the problem.

2

If it's highly correlated with the outcome and not correlated with the other predictors, you should almost certainly include it, as it increases the power to detect the effect of other other predictors. In a randomized trial, the baseline measure of the outcome variable is the most important control variable. If it is correlated with the other predictors, ...

2

Firstly: My calculations yield a required sample size of $239$ (see below for more details). Secondly: In this context, the concept of power is not needed at all. Power is only needed in the context of statistical testing, where you need to know the distribution of your test statistic under $H_1$. In this context, you are simply interested in "sharpening" ...

2

Autocorrelation is only meaningful when the data is ordered, such as in time series that are naturally ordered along the time scale, or when the distance between the observations is meaningful, such as the case of spatial data. Edit: Another case where the distance between the observations is meaningful is highlighted by @whuber's comment: it may reflect the ...

2

Matching creates complexity in analyses, which is why you see many researchers actually ignore the matching when the analysis is done. Not good. You can see from your question that matching creates the need to make a lot of arbitrary but far-reaching decisions. Not good. And any method that throws away valid observations is not very good from a purely ...

2

No, I recommend against that. Correlation coefficients are not on an interval scale, so summing, say, .5+.5 will reflect something different from summing .2 + .8, much less 0+1.0. If you want to do this, computing Fisher's z first would help it make more sense. If you want to keep it in correlation metric, you could average the zs (instead of summing) and ...

2

A general advise would be going online on university statistics department websites and on individual class pages and looking at their recommended textbooks. Here is the graduate level probability class website from UC Berkeley: https://www.stat.berkeley.edu/~aldous/205A/index.html Here is the graduate level statistics class website from UC Berkeley: https:...

2

There is probably some built-in way to do it, but you can always do linear regression step by step with matrices. If your dependent variable is a column $Y$ and your independent variables are a set of columns $X$, the vector of regression parameters is: $(X'X)^{-1}X'Y$ So first you would use the TRANSPOSE function to transpose matrix $X$, which will give ...

Only top voted, non community-wiki answers of a minimum length are eligible